IET Research Journals

Enhanced Load Frequency ControlIncorporating Locational Information forTemporal Enhancement

ISSN 1751-8644doi 0000000000wwwietdlorg

Mazheruddin H Syed1 Efren Guillo-Sansano1 Steven M Blair1 Graeme M Burt1 Alexander MProstejovsky2 and Evangelos Rikos31 Institute for Energy and Environment University of Strathlcyde Glasgow G1 1RD UK2Department of Electrical Engineering Technical University of Denmark Riso 4000 Denmark3Center for Renewable Energy Sources and Saving Pikermi Attiki 19009 Greece E-mail mazheruddinsyedstrathacuk

Abstract With the increasing penetration of renewables in power systems frequency regulation is proving to be a major challengefor system operators using slower conventional generation units and alternative means to provide faster regulation are beingactively sought The participation of demand side management in ancillary service provision is proven in a number of energymarkets yet its full potential to benefit frequency regulation including the exploitation of fast power ramping capability of somedevices is still undergoing research In this paper a novel approach to improve the speed of response of load frequency controla secondary frequency control approach is proposed The proposed control is enabled by an effective location identificationtechnique is highly resilient to anticipated system changes such as reduction of inertia and enables fully decentralized powersystem architectures The effectiveness of the approach is demonstrated and compared to that of present day regulation controlby means of real-time simulations incorporating appropriate time delays conducted on a five-area reduced model of the GreatBritain power system The applicability of the method is further proven under realistic communications delays and measurementsexperimentally using a controller and power hardware-in-the-loop setup demonstrating its critical support for enabling the stableoperation of future power systems

Nomenclature

Textual AbbreviationsACE area control errorADD area disturbance determinerBCL balance control loopBESS battery energy storage systemCG conventional generationCLFC conventional load frequency controlDSA demand side aggregatorELFC enhanced load frequency controlFCDM frequency control by demand managementGB Great BritainLFC load frequency controlPFC primary frequency controlPI proportional integralRoCoF rate of change of frequencySFC secondary frequency controlMathematical FormulationsαDSAi αCGi participation factors for demand side aggregators

and conventional generators respectivelyβi frequency bias factor for area iδij area i and j tie-line breaker state∆fi change in frequency of area i∆fovr relative frequency overshoot∆P power imbalance∆PBCL

i control effort of auxiliary controller∆Pci control effort of secondary control∆PDSAlowast

i demand side aggregator power regulation∆Pgi deviation in generator valve position∆PLi load power variation for area i∆Pmi deviation in generator mechanical output∆P tie

i tie-line power deviation for area i∆P tie

Thi tie-line power deviation threshold

∆P tielowasti filtered tie-line power deviation

ε error marginmicrostep microinertia step size and inertia multipliers respectivelymicro σ mean value and standard deviation respectivelyΣ∆PDSA

i demand side aggregator power variation for area iA synchronous areadfdt df

dt

lowast dfdt

lowastlowastrate of change of frequency measured extremumand filtered respectively

di output of area disturbance determinerDi load damping coefficient for area ifdb frequency dead-bandfnom fmeas nominal and measured frequency respectivelyfmax fmin maximum and minimum frequency respectivelyHi H0 inertia constant for area i and reference inertia of

the system respectivelyi index representing area under considerationj index representing adjacent areas to area indexed ik scaling factorKPi TIi proportional gain and integral time constant of

secondary controller for area i respectivelyKPaux TIaux proportional gain and integral time constant of

auxiliary controller for area i respectivelyM number of control areasNHi neighborhood of areaP step P step

0 disturbance size and reference disturbance sizeP tie0i P tie

measi scheduled and measured tie-line power flowRi speed droop for area iTai Tbi turbine time constants for area i respectivelyTgi governor time constant for area ith tk control time step and a chosen time instantTf filter time constantTn measurement window lengthT rest frequency restoration timeuw x control disturbance and state vectors respectively

IET Research Journals pp 1ndash10ccopy The Institution of Engineering and Technology 2015 1

1 Introduction

Increasing penetration of renewable generation sources the decreas-ing population of conventional synchronous generation (CGs) andthe consequent decrease in system inertia deteriorates the systemfrequency response In a study undertaken by the Australian EnergyMarket Operator it was shown that with the forecasted change ingeneration mix (ie a rapid uptake of non-synchronous generation)the rate of change of frequency (RoCoF) levels will increase suchthat by 2022 there would be less than two seconds for primary fre-quency control (PFC) actions to arrest frequency before it declinesbelow the defined containment band [1] The two seconds periodis much shorter than the response time typically observed for CGsIt can therefore be said that the decreasing inertia in the networkimposes new requirements on existing services In recent literaturea number of works have proposed novel synthetic-inertial controlapproaches [2ndash5] to alleviate the problem of declining system iner-tia however in practice synthetic inertia is commercially unavail-able [6] Furthermore as reported in [6] virtual inertia emulationis feasible within power networks with high inertia but it becomesincreasingly challenging to obtain an appropriate transient responsein low inertia circumstances Owing to the adaptableflexible natureof power electronic interfaced devices they can yield a wider rangeof responses and enforcing only inertial response from such devicescan be limiting and can curb further developments Other systemoperators around the world have noted this common concern ofdeclining system inertia and its effect on system frequency and thechallenges brought forwards towards frequency control [7 8] Real-ising the need for faster acting service provision system operatorsare working towards better understanding of the impact and inte-gration of fast response in the entire spectrum of frequency controlmechanisms (such as inertial PFC secondary frequency control) asis evident from the pilots and plans of many international systemoperators summarized in [9]

Load frequency control (LFC) also referred to as automaticgeneration control is a secondary frequency control (SFC) pro-cess employed to maintain the frequency of interconnected power-systems such as in Continental Europe USA and Australia [10ndash12]LFC is relatively slow designed for responding only after the fre-quency has been contained by the PFC The performance of LFC canconventionally be improved by the use of fast acting gas turbines asthe regulation providing units however their availability and cost ofoperation can restrict their implementation [13] An alternative to theconventional approach is the use of flexible demand available withinthe network In [14] and [15] it has been shown that the utilization ofa utility scale battery energy storage system (BESS) in conjunctionwith CGs improves the LFC delivering performance of the CGs In astudy that characterizes the effectiveness of utilizing battery energystorage systems for LFC compared to CGs a reduction in the reg-ulation requirement is observed even when they constitute a smallportion of total capacity [16] This has also been confirmed in studiesconducted by [17] Although the performance of LFC is improved byutilizing fast acting devices the slow operation design of LFC dueto ramp rate constraints on CGs does not allow for exploitation ofthe full potential of such devices

An approach already in use in the Great Britain (GB) power sys-tem and referred to as frequency control by demand management(FCDM) [18] provides an alternative means to exploit the powerramping potential of fast acting devices This approach is basedon determining a frequency threshold for participating fast actingdevices [19] The amount of reserves activated by such means issubtracted from the LFC regulation command In [20] it was shownthat such an approach alleviates the burden on LFC and provides abetter frequency response At present only relatively large commer-cial customers participate in FCDM and are activated by means ofunder-frequency relays The dimensioning of FCDM reserves andthe thresholds for their activation are carefully determined basedon the system power frequency characteristics In the future withhigher penetration of intermittent renewable energy resources withinthe grid this reserve dimensioning and thresholds determination willprove to be even more challenging as highly variable system power

frequency characteristics are anticipated The opportunity of partic-ipating in FCDM will be extended to the large amount of flexibilityforeseen to be available within the grid (including domestic devicessuch as electric vehicles and heat pumps) Devices interfaced bypower electronics will have no difficulty in participating throughlocal measurements however some devices that inherently do notincorporate local observables may not be able to participate In otherwords the large amount of distributed domestic flexibility withinthe network will be available to participate through the traditionalLFC (as it requires communications enablement rather than localmeasurements)

In [21ndash23] alternative structures for LFC and controls for theparticipation of fast acting devices have been proposed The par-ticipation of demand side fast acting devices in LFC structureshas further been facilitated by the establishment of demand sideaggregators (DSA) that can be perceived as any other regulatingreserve-providing unit under LFC [24]

To summarize it can be said that in the past there had been noincentive to improve the speed of response of SFC With the changesexpected within the power system in near future there is a growinginterest in improving the speed of response of frequency control ser-vices including SFC Improving the response speed of SFC (i) islimited due to the ramp rate constraints on participating CGs and (ii)can deteriorate the system frequency response due to SFC interactionwith the PFC response As discussed above even with the increas-ing presence of fast acting devices within the network the speed ofresponse of SFC within the conventional architecture of LFC cannotbe improved

To address the above gaps in this paper a novel SFC is proposedThe proposed control incorporates locational information ie thelocation of the disturbance (power imbalance due to lossadditionof generation or load is considered a disturbance in this paper) toenable the temporal enhancement of LFC The proposed controlcomprises a fast acting balance control loop (BCL) in addition tothe secondary control loop of LFC The BCL comprises an areadisturbance determiner capable of fast autonomous and definitiveidentification of disturbance location The fast identification of thedisturbance location allows for unilateral activation of reserves onlywithin the area where the disturbance has occurred at increasedspeeds which mitigates the deteriorating impact of its interactionwith PFC In addition the control effort to fast acting resourcesis a combination of effort from BCL and secondary control loopwhile the control effort to slower acting resources is from secondarycontrol loop only thereby eliminating the restrictions on ramp ratesconventionally imposed The proposed control is simple and doesnot require additional forecasting or complete network awarenessthereby allowing seamless integration within present day LFC Theperformance of the proposed control is compared with that of con-ventional LFC within a reduced five-area dynamic model of theGreat Britain power system by real-time simulations corroboratedanalytically by small-signal analysis and its real world applicabilityproven by experimental evaluation within a laboratory

2 Proposed Frequency Control Approach

In this section to present the proposed approach effectively the con-ventional LFC (referred to as CLFC henceforth) is first introducedand its shortcomings highlighted This is followed by the descriptionof the proposed control approach

21 Conventional Load Frequency Control (CLFC)

Consider an interconnected power system with M control areasindexed by i = 1 2 middot middot middot M The CLFC model of the ith controlarea is presented in Fig 1 As shown the CLFC comprises a primarycontrol loop and a secondary control loop where the aim of primarycontrol is to contain the frequency deviation caused by power imbal-ance in any control area while the secondary control is responsibleto recover the frequency back to its nominal value [5] The sys-tem dynamics of the ith area can be represented by the following

IET Research Journals pp 1ndash102 ccopy The Institution of Engineering and Technology 2015

Fig 1 Conventional load frequency control (CLFC)

differential equations

(1)∆fi (t) = minus D

2Hi∆fi (t) +

1

2Hi

(∆Pmi (t) + ∆P ren

i (t)

minus∆PLi (t)minus∆P tiei (t) + Σ∆PDSA

i (t))

(2)∆Pmi (t) = minus 1

Tbi∆Pmi (t) +

1

Tbi∆Pgi(t) +

TaiTbi

∆ ˙Pgi (t)

∆ ˙Pgi (t) = minus 1

Tgi∆Pgi (t) +

1

TgiαCGi∆Pci (t)minus 1

RiT gi∆fi (t)

(3)

where

(4)∆P tiei =

2π

s

Msumj=1j 6=i

Tij(∆fi minus∆fj

)where ∆fi is the change in frequency ∆Pci denotes control effortof secondary control ∆Pmi and ∆Pgi are the deviations of gen-erator mechanical output and valve position for area i respectivelyHi Di Ri Tgi Tai and Tbi are the inertia constant load damp-ing coefficient speed droop governor and turbine time constants forarea i respectively ∆P ren

i ∆PLi ∆P tiei and ∆PDSA

i are powervariations of renewable generation loads tie-line and DSAs respec-tively which can be viewed as external disturbances to the system∆fj is the change in frequency of area j and Tij is the synchroniz-ing torque The secondary control input is a signal referred to as thearea control error (ACE) and can be defined as

(5)ACEi (t) = βi∆fi (t) + ∆P tiei (t)

where βi is the frequency bias factor Upon occurrence of an eventthat leads to a deviation in frequency and ACE beyond a set thresh-old a proportional integral (PI) control is employed to force the ACEto zero and can be represented as

(6)∆Pci (t) = minusKPiACEi (t)minus 1

TIi

intACEi (t)

whereKPi TIi are the proportional gain and integral time constantrespectively

22 Shortcomings of Conventional Load Frequency Control

Theoretically the distinction between primary and secondary fre-quency control is their control objective ie primary containsfrequency deviations and secondary restores the frequency to itsnominal value Practically a time scale of operation is also asso-ciated with the two controls ie the primary response is expectedwithin ten seconds of the observed disturbance while secondaryresponse can vary from thirty seconds to minutes This is in gen-eral regarded as the secondary control intentionally being designed

Fig 2 Enhanced load frequency control (ELFC)

for much slower operation than primary control However this asso-ciation of time with control objective is due to practical limitationsas identified below

1 Slow disturbance location identification Within CLFC althoughthe entire synchronous area perceives the disturbance as a changein frequency only the area with the disturbance is responsible torestore the frequency This decentralized operation is achieved bymeans of ACE ie assuming appropriate calculation of β the ACEdistinguishes an internal disturbance (within the area) from an exter-nal one (outside the area) A definitive identification can only beobtained when the RoCoF df

dt = 0 and hence the delay in the oper-ation of the secondary control

2 Transient ACE following a disturbance Following a disturbancethe ACE of non-disturbance areas slowly converges to zero If thespeed of response of secondary control was increased the transientnon-zero ACE would enforce unnecessary secondary participationfrom non-disturbance areas This can further be exacerbated when βis miscalculated

3 Ramp-rate constraint of CGs Large CGs constitute the major-ity of the secondary reserves The ramp rate constraint of theseparticipating generators limits the speed of response of secondarycontrol

23 Proposed Enhanced Load Frequency Control (ELFC)

To overcome the limitations of the conventional approach a novelenhanced LFC (ELFC) as illustrated in Fig 2 is proposed The con-trol incorporates a fast acting BCL in addition to the primary andsecondary control loops The key features of the proposed approachare as follows

bull The BCL comprises an area disturbance determiner (ADD) capa-ble of fast and autonomous identification of disturbance location in acompletely decentralized manner This ensures activation of reservesunilaterally only within the area that has initiated the disturbanceallowing for faster response speeds (faster reserve activations)

bull The area power imbalance observed over tie-lines (∆P tiei ) is the

control input of BCL as opposed to the conventionally utilized ACEThe non dependence of control input on frequency eliminates dete-riorating interaction with primary control and avoids unnecessaryactivations

bull The BCL employs only fast acting demand side devices throughDSAs as regulation providing reserves eliminating any restrictionon speed of response due to ramp-rate constraints Therefore thepower regulation command of the DSAs comprises two parts (i)regulation command from secondary control loop with predefinedparticipation factor αDSAi and (ii) regulation command from theBCL

The fast acting nature of the BCL enables active contribution toobjectives of both primary and secondary control hence justify-ing its existence as an independent loop The description of the fastacting BCL is presented in the following section

IET Research Journals pp 1ndash10ccopy The Institution of Engineering and Technology 2015 3

3 Fast Acting Balance Control Loop (BCL)

In this section the operation of the fast acting BCL is explained

31 Fast Location Identification by Area DisturbanceDeterminer (ADD)

The objective of the ADD is to autonomously quickly and unam-biguously distinguish between disturbances originating within thearea and those originating elsewhere in the synchronous region Theoperating principle of the ADD is presented below

Any sudden imbalance between generator input mechanicalpower and load will lead to a deviation in system frequency thedynamics of which can be represented by the swing equation as

∆P =2H

fnom

df

dt(7)

where ∆P is the power imbalance in the system H is the inertiaconstant of the system fnom is the nominal frequency and df

dt is theRoCoF RoCoF at a specific time instant tk can be estimated as

df

dt=ftk minus ftkTn

Tn(8)

where Tn is the length of the measuring window Within a syn-chronous area A and for area under consideration i the neighbor-hood can be defined as NHi = iAN i sube A With j isin AN ias adjacent areas coupled over tie-lines with breaker state δij thechange in tie-line power flow for area i subject to the disturbancecan be calculated as

(9)∆P tiei =

sumjisinAN i

δijPtie0i minus

sumjisinAN i

δijPtiemeasi

where P tie0i and P tie

measi are the scheduled and measured tie-linepower flows respectively As the change in electric power flow isimmediate upon occurrence of power imbalance such that ∆P tie

i ge∆P tie

Thi measuring the extremum of dfdt as df

dt

lowast for ∆P tie

i dfdtlowast6=

0 the disturbance is classified as originating within the area if

(10)

dfdt

lowast∣∣∣dfdt lowast∣∣∣+ 1

minus minusdf

dt

lowast∣∣∣dfdt lowast∣∣∣+ 1

6=

lfloor∆P tie

i∣∣∆P tiei

∣∣+ 1

rfloorminus

lfloorminus∆P tie

i∣∣∆P tiei

∣∣+ 1

rfloor

where ∆P tieThi is the threshold of change in tie-line power flow set

based on the largest expected load variation in area i It should benoted that Eq (10) is an alternative form of signum function usingfloor (b c) and absolute (| |) functions Such representation is moreappropriate in this case due to its use for evaluation of real numbersexcept zero ie for ∆P tie

i dfdtlowast6= 0 An alternative representation

for convenience of implementation in embedded platforms is withthe use of Iversons brackets [25] as minus

[dfdt

lowastlt 0]

+[dfdt

lowastgt 0]6=

minus[∆P tie

i lt 0]

+[∆P tie

i gt 0] Upon satisfying Eq (10) ie the

sign of ∆P tiei is not equal to the sign of df

dt

lowast the output of the

disturbance determiner is di = ∆P tiei indicating the disturbance to

be within the area else di = 0 indicating the disturbance is out-with the area Therefore each area of A can identify disturbancesoriginating within themselves based on local measurements in adecentralized manner Although the identification is fast there mightbe a frequency dead-band (fdb) such that for measured frequencyfmeas no frequency control measure is employed when minusfdb ltfmeas minus fnom lt fdb

32 Auxiliary Controller

The output of ADD is an input to the auxiliary controller and forwhen Eq (10) is true di should be controlled to zero Employing aPI control the control effort can be represented as

∆PBCLi (t) = minusKPauxdi (t)minus

1

TIaux

intdi (t) (11)

where Kpaux TIaux are the proportional gain and integral timeconstant of the auxiliary controller respectively Contrary to the sec-ondary control loop where the value of TI is limited by the ramprate constraints of CGs TIaux can be chosen to achieve a fasterresponse The use of PI control for BCL is not mandatory and theoutput of ADD can be an input for direct load frequency control aspresented in [26] The effective power regulation command issued tothe DSAs can be represented as

∆PDSAlowast

i (t) = ∆PBCLi (t) + αDSAi∆Pci (t) (12)

33 Low Pass Filtering

To suppress the measurement noise and to smoothen the responseof BCL a low pass filter is utilized for ∆P tie

i and dfdt The filtered

signals at time instant tk can be represented as

∆P tielowasti (tk) = (1minus a) ∆P tielowast

i (tk minus 1) + a∆P tiei (tk) (13)

df

dt

lowastlowast(tk) = (1minus a)

df

dt

lowastlowast(tk minus 1) + a

df

dt(tk) (14)

where a = thTf+th

with th as the time step of control implementa-tion and Tf as the filter time constant It is important to choose thefilter time constant such that it renders satisfactory quality of inputwhile attributing least delay

4 Test System Characterization

In this paper the CLFC is the reference implementation againstwhich the performance of the proposed ELFC will be assessed In thefollowing sub-sections the test system characteristics are detailed

Fig 3 Reference five-area reduced GB power system

IET Research Journals pp 1ndash104 ccopy The Institution of Engineering and Technology 2015

41 Test Power System

For analyzing the proposed frequency control approach in this papera reduced six-machine dynamic model of the GB power system hasbeen chosen as the test grid It should be noted that the GB gridoperates as one synchronous area However in this paper the busesof the reduced model have been grouped to represent multiple LFCareas as shown in Fig 3 each of which relates to one or moreregions identified in the GB National Electricity Transmission Sys-tem boundary map presented in the Electricity Ten Year Statementof the Transmission System Operator [27]

411 Modeling The model has been developed in RSCAD andsimulated in real-time using a digital real-time simulator from RTDSTechnologies with each area comprising at least one aggregatedgenerator and an aggregated load Each aggregated generator is mod-eled as a large synchronous machine connected to the transmissionsystem via a step-up transformer (138400kV) The rating of eachgenerator is set according to a characteristic GB 2017 load flow[27] Each generator is controlled by the widely used IEEE type1 static excitation system A gas turbine and speed governor con-trol the speed and input torque of each machine The synchronousmachine excitation system and gas turbine parameters have beenobtained from [28] while the governor speed control parameters aretuned against real recorded events as will be explained in the follow-ing sub-section The transmission lines are modeled using π-sectionswith lumped resistance capacitance and inductance parameters cal-culated from the power flow data provided by the Electricity TenYear Statement for 2017 [27] as per the methodology presented in[29] The model parameters can be found in [30] and hence havenot been repeated in this paper

412 Validation The model has been validated by means oftwo tests [30]

1 Load flow analysis This test is to evaluate the steady-state per-formance of the model The load flow simulation data of the testsystem closely matches the winter peak 2017 data [27] This includesthe generation and demand data at each region and the power flowacross the boundaries of the regions

2 Dynamic frequency response evaluation The dynamic responseof the reference power system model is benchmarked against a mul-tiple real historic frequency deviation events with data recordedby phasor measurement units located at various points of the GBpower transmission network In [30] the event chosen was the tripof the England-France high voltage direct current inter-connectoron 11th of January 2016 leading to a power loss of 900 MW Inother words the model represents the real-world GB network onthe 11th of January 2016 The total generation and demand of themodel is adjusted to match the values on the day of the event (totaldemand=5956GW) The inertia constant the governor time con-stant the droop percentage and the load reference set-point param-eters are tuned to ensure the model frequency response matches thefrequency response (pre-disturbance RoCoF and frequency nadir)obtained from the phasor measurement units

42 Control Implementation

The following controls are implemented in each of the five areas ofthe GB power system with nominal frequency fnom = 50Hz Theparameters utilized are presented below

1 Primary Control Loop The synchronous generators in each ofthe areas participate proportionally to their capacity to contain thefrequency with a droop setting of 13

2 Secondary Control Loop The typical values of KPi and TIi arein the range of 01minus 10 and 50minus 200s respectively [31] [32] Thebest (fastest) value of TI = 50s has been chosen The value of KPihas been chosen as 01 Both the synchronous generator and aggre-gated load constitute the secondary reserves with αCGi = 025 and

αDSAi = 075 while fdb is selected as 01Hz [28] [33] [11]

3 Balance Control Loop With aggregated loads participating asreserves the values of KPaux and TIaux are selected as 01 and5s respectively With no ramp-rate constraints a smaller value ofTIaux has been chosen to demonstrate the effectiveness of BCLFor the ADD the values of ∆P tie

Thi = 50MW Tn = 002s andTf = 002s are selected

4 Delays Measurement and communication delays are incorpo-rated within the secondary and balance control loop The values ofACE and ∆P tie are thus updated every 2s representing the measure-ment delay [34] while the outputs of the two control loops ∆Pci

and ∆PDSAi are delayed by 1s representing communications delay

[35]

43 Incorporation of Renewable Generation Penetration

To analyze the performance and aptness of the proposed frequencycontrol approach in a more representative power system the incor-poration of renewable generation within the test power system isnecessary In this sub-section the incorporation of the impact ofrenewable generation penetration in particular the reduction in iner-tia within the test power system is explained First a multipliermicroinertia is defined that scales the system inertia constant (H) suchthat H = microinertiaH0 where H0 is the reference system inertia Toelaborate for example a value of microinertia = 09 leads to a 10reduction in system inertia Accordingly the inertia constant andpower generation of each of the six-machines within the test powersystem is reduced by 10 As sub-cycle phenomena are not underconsideration a simpler first-order equivalent model representingnon-synchronous generation [36] in each of the areas provide theremainder of the power generation

44 Demand Side Aggregators (DSAs)

DSAs in this work correspond to energy management systems foractive residential customers andor commercial buildings (referred toas prosumers) with fast-acting responsive loads modeled as dynamicloads in simulation The response dynamics of DSAs ie theresponse delay and uncertainties are subject to accurate load mod-eling and capacity estimation that are by nature time-varying andan important field of research on its own Two types of delays con-tribute towards the response delay of DSAs the communicationdelays between the DSA and the participating prosumer load andthe response time (onoff delay) of the prosumer load upon receiptof activation command from DSA The response characteristics of aprosumer load such as the minimum time a load has to stay onoffafter activation or the minimum time before the same load can becalled on for a second activation are considered a part of DSAcapacity estimation that is assumed to be guaranteed in principleie either by regulation or penalty mechanisms enforced by DSAitself However the communications delay is network-dependentand should be considered for establishing the performance of fre-quency control mechanisms Therefore in this paper it is assumedthat the required amount of restoration reserves are procured byeach of the LFC areas while the participating loads are assumed torespond to the control command immediately with no turn onoffdelays [37] considered for this work

45 Other Characteristics

A number of occurrences of sim1GW generation loss have beenexperienced by the GB grid within the last year and therefore adisturbance magnitude of 1GW within area 2 has been selected asthe reference frequency disturbance [38] The net generation loss isemulated by a step increase in load and a net load loss by a stepdecrease

IET Research Journals pp 1ndash10ccopy The Institution of Engineering and Technology 2015 5

5 Performance Evaluation

In this section the applicability of the proposed ELFC is evaluatedby means of real-time simulations and experiments within a state-of-the-art smart grid laboratory A three step approach for validation isadopted as follows

1 Simulation and analysis The feasibility of the approach is veri-fied by means of real-time simulations The response of the systemsubject to reference disturbance is evaluated followed by a sensitiv-ity analysis with respect to system inertia and size of disturbanceThis is followed by bringing forth the advantages offered by the keyfeatures of the BCL ie the ADD and use of tie-line power flow(∆P tie

i ) as control input instead of the ACE2 Controller Hardware-in-the-Loop validation Such a validationserves two main purposes (i) the validation of the controller underrealistic delays and (ii) the prototyping of the proposed controlwithin an actual hardware controller to demonstrate its real-worldapplication3 Controller and Power Hardware-in-the-Loop validation The per-formance improvement of the proposed ELFC over CLFC has beenachieved by means of incorporation of BCL that relies on ADDfor a faster disturbance detection The disturbance detection utilizesRoCoF that is regarded as a very noisy measurement Any con-trol that relies on RoCoF should therefore be validated by means ofreal measurements enabled by controller and power hardware-in-the-loop implementation Therefore the ADD of the proposed control isvalidated in this way

In all the above cases the performance of the proposed ELFC iscompared to that of CLFC To aid the assessment two key indica-tors are defined namely (i) frequency restoration time (T rest) and(ii) relative frequency overshoot (∆fovr) To asses the T rest of thecontrollers an error margin ε = 001 is defined such that T rest isthe time interval between the initiation of the disturbance to the pointwhen |∆f |= |fmeas minus fnom|lt ε and can be represented as

T rest = T rest isin R forall t gt T rest |∆f |le ε (15)

Defining the minimum and maximum frequency after a disturbanceas fmax and fmin respectively the relative overshoot is calculatedas

∆fovr =

∣∣∣∣fmax minus fnomfnom minus fmin

∣∣∣∣ (16)

As the GB power system is a relatively strongly coupled networkie a stiff network therefore the same frequency characteristics areobserved throughout the network [39] For simplicity and clarityof the figures only the frequency of the area with a disturbance isplotted for the analysis that follows

51 Simulation and analysis

511 Response to reference disturbance The system fre-quency responses subject to a generation loss at t = 10s and a loadloss at t = 110s for CLFC and ELFC are presented in Fig 4a Therestoration time for the CLFC is 83s with no overshoot The pro-posed ELFC by means of fast and accurate detection of disturbancelocation contributes to improving frequency nadir (frequency zenithfor load loss) and leads to a faster restoration time With a restora-tion time of 352s the proposed ELFC is twice as fast but presentsan overshoot of 35 An acceptable frequency response of a sys-tem subject to a disturbance is defined by an exponentially decayingfunction H(s) = fnom plusmnAeminus

1T t referred to as the trumpet curve

set as a requirement by the European Network of Transmission Sys-tem Operators for Electricity [31] Considering the response of theconventional control as reference trumpet curve parameters are cal-culated as A = 031Hz and T = 11s As can be observed from Fig4a the response of the proposed ELFC with a 35 overshoot iswell within the bound set by the trumpet curve The overshoot of theELFC can be abated and the response further improved by means

0 50 100 150 200

Time (s)

497

498

499

50

501

502

503

Fre

qu

en

cy

(H

z)

ELFC

CLFC

Trumpet Curve

Trumpet Curve

(a) Frequency response

0 50 100 150 200-005

0

005

AD

D O

utp

ut

Pti

e0 50 100 150 200

Time (s)

0

002

004

006

To

tal

Re

gu

lati

on

Po

we

r (p

u)

10 101 102 103 104

-004

0

Other LFC

ADD Outputs

LFC 2 ADD

Output

LFC 2

Other

LFCs

(b) ADD output and total regulation power for ELFC

microStep

microInertia

701

80

09 2

Re

sto

rati

on

Tim

e (

s)

90

CLFC

08 15

100

071

06 microStep

microInertia

25

1

30

09 2

Re

sto

rati

on

Tim

e (

s)

35

ELFC

08 15

40

071

0630

40

50

60

70

80

90

100

(c) Sensitivity analysis

Fig 4 Simulation and sensitivity analysis results for performanceevaluation of proposed ELFC

of faster acting PFC where primary response is provided by bat-tery energy storage systems as opposed to CGs as is the case forsimulations undertaken

The output of ADD and the total regulation power ie the sumof the regulation command from the DSAs and the secondary con-trol loop for all the areas subject to the generation loss and loadloss is shown in Fig 4b As can be observed only the output ofLFC 2 ADD is non-zero (and equal to ∆P tie

2 ) consequent to suc-cessful disturbance identification Furthermore this identification iswithin 300ms (sim 250ms) as shown This enables the fast activationof resources through the BCL to be activated unilaterally ie only inarea with disturbance as is evident from the total regulation powershown

512 Sensitivity analysis To further evaluate the performanceof the proposed ELFC a sensitivity analysis for variation in size ofdisturbance and system inertia is undertaken The size of disturbance(P step) is varied by means of a multiplier microstep such that P step =

microstepPstep0 where P step

0 is the reference disturbance magnitudeIn a similar manner a multiplier microinertia scales the system inertiaconstant (H) such that H = microinertiaH0 where H0 = 468s Theresults of the analysis for CLFC and ELFC are presented in Fig 4cAs can be observed the restoration time for CLFC increases with

IET Research Journals pp 1ndash106 ccopy The Institution of Engineering and Technology 2015

(a) CLFC with auxiliary controller-002 0 002 006

-100

-50

0

50

100

Imag

inary

Axis

(seco

nd

s-1

)

(seconds-1)

(seconds

-1)

-003 -001 0 001

-100

-50

0

50

100

Imag

inary

Axis

(seco

nd

s-1

)

(seconds-1)

(seconds

-1)

CLFC ELFC

Real Axis (seconds )-1Real Axis (seconds )-1

TIaux

=50 s

TIaux

=5 s

TIaux

=50 s

TIaux

=5 s TIaux

= 50 s rarr 5 s

TIaux

= 50 s rarr 5 s

(b) Eigenvalue analysis

0 10 20 30 40 50 60 70 80 90 100

Time (s)

497

498

499

50

501

502

Fre

qu

en

cy (

Hz)

Accurate Bias Factor

5-

Error in Bias Factor

5+

Error in Bias Factor

ELFC

Responses

CLFC

Responses

(c) Frequency response

0 10 20 30 40 50 60 70 80 90 100

Time (s)

0

002

004

006

008

01

Reg

ula

tio

n P

ow

er

(pu

)

5-

Error in Bias Factor

5+

Error in Bias Factor

Responses for

ELFC using ACE

Responses for

ELFC using ∆ Ptie

(d) Regulation power

Fig 5 Results for performance evaluation of proposed ELFC

increasing disturbance size but the decreasing system inertia doesnot have a significant impact On the other hand the performanceof proposed ELFC is significantly different the restoration time ofwhich is not affected by the increase in disturbance size while atthe same time a decrease in system inertia leads to faster restora-tion times Therefore the proposed ELFC is more robust to varyingdisturbance sizes and system characteristics a necessity for futurepower systems

With variations in disturbance size and system inertia the relativeovershoot for the proposed ELFC is always under 40 and withinthe bounds set by the respective trumpet curve While the CLFCexhibits no overshooting characteristic with decreasing system iner-tia the frequency nadir significantly decreases The proposed ELFCimproves the frequency nadir for all the cases assessed

513 Importance of ADD The objective of this sub-section isto demonstrate that the improvement in the performance of the pro-posed control is not only due to the incorporation of additional fasterauxiliary controller but the fact that the identification of the locationof the disturbance does play a very important role Therefore theimportance of ADD is demonstrated by means of conducting small-signal stability analysis on the test power system The two cases forcomparison are (i) the proposed ELFC control and (ii) the CLFCincorporating the auxiliary controller only as shown in Fig 5a Thestate-space model of M-area interconnected power system with BCL(as shown in Fig 2) can be represented as [40]

x = Ax+Bu+ Fw

y = Cx(17)

with x = [x1 x2 x3 x4 x5]T

u = [u1 u2]T

w = [∆PL1 ∆PLM

]T(18)

where x is the state vector u is the control vector andw is the disturbance vector The internal states can be rep-resented as x1 = [∆f1 ∆fM ] x2 = [∆Pm1 ∆PmM ]x3 =

[∆P tie

1 ∆P tieM

] x4 =

[intACE1

intACEM

]and

x5 =[intd1

intdM] The internal inputs can be represented

as u1 = [∆Pc1 ∆PcM ] and u2 =[∆PDSA

1 ∆PDSAM

]

The coefficient matrices A B and C can be found in [40] Thesystem is linearized around fnom in steady-state The input to thesystem is the reference disturbance in area 2 and the output is thesystem frequency

The Eigenvalues for the two cases are obtained by varying theTIaux from 50s to 20s in steps of 10s and then to 5s in steps of 5sIn Fig 5b the pole-zero map for the CLFC (left) and the proposedELFC (right) have been presented Only the states that are impactedby TIaux are shown for clearer representation As can be observedfrom Fig 5b for CLFC as the value of TIaux is reduced the polesmove towards the imaginary axis For TIaux = 10s the poles crossthe imaginary axis representing an unstable system However forthe proposed ELFC where the ADD ensures unilateral reserve acti-vations only within the area where the disturbance has occurred thepoles are very slightly impacted by the reduction in TIaux whichtherefore does not deteriorate the stability of the system

514 Choosing ∆P tie over ACE This sub-section presents ajustification for the use of ∆P tie within BCL as opposed to the con-ventionally used ACE Consider the frequency responses presentedin Fig 5c with frequency bias factor miscalculation of 5 for boththe CLFC and proposed ELFC As can be observed a 5 miscalcu-lation of β results in a plusmn17s variation in T set for the CLFC whilehas no significant impact within the proposed ELFC β is a combina-tion of the droop set for the area and the damping offered by the loadwithin the area and it is not uncommon for errors in its calculationLarger generating units within the network are bound by the gridcode to provide a set percentage response to a variation in frequencyHowever if the units do not abide by the set droop this will lead tomiscalculation of β Furthermore presently it is normal practice todetermine β on a yearly basis [11] In future systems it is expectedthat the characteristics of the network (eg system inertia) will varyvastly even within a single day It is also expected that the differencebetween the peak load and the base load will increase significantly[7] thereby resulting in the load offering a different damping effectat different times Therefore the proposed ELFC utilizing ∆P tie ismore resilient to changes expected within the future power system

IET Research Journals pp 1ndash10ccopy The Institution of Engineering and Technology 2015 7

1 15 2 25 3 35 4 45

Delay (s)

0

1000

2000

3000

4000

5000

Sam

ple

s

Uniform Distribution

Gaussian Distribution

(a) Delay distributions under consideration

(b) Controller hardware-in-the-loop setup

0 10 20 30 40 50 60 70 80 90 100

Time (s)

4975

498

4985

499

4995

50

5005

501

Fre

qu

en

cy (

Hz)

No delay

Gaussian Distribution Delay

Uniform Distribution Delay

(c) Frequency response with and without delays

Fig 6 Setup and results for controller hardware-in-the-loop imple-mentation

Furthermore if ACE was instead utilized within the BCL of theproposed scheme although the activations of reserves are guaran-teed to be unilateral within the area where the disturbance has beendetected the miscalculation of bias factor would lead to nuisanceactivations as shown in Fig 5d It should be noted that utilizing∆P tie avoids any nuisance activations

52 Controller Hardware-in-the-Loop Validation

A detailed study on experimental validation of ancillary serviceprovision capability of DSAs controlling a portfolio of highly dis-tributed demand side devices has been undertaken in [41] Thestudy identifies that validation of any control relying on DSAs forprovision of time critical ancillary services should incorporate real-istic communication delays to demonstrate robustness The resultspresented in previous sub-sections assume a fixed communicationsdelay however the communication delay referred to in this paperis the time it takes for a control command from the aggregator toreach the participating device and is highly dependent upon the com-munications architecture employed by the potential demand sideaggregator

The performance of the communications networks that connect acontrol centre to the end devices for demand side applications hasbeen analyzed in [42] and the results show that the latency expected

Table 1 Test Metrics

Delay TRest (s) ∆fovr ()

No delay 2416 287Gaussian distribution delay 4237 44Uniform distribution delay 423 436

(a) controller and power hardware-in-the-loop setup

0 10 20 30 40 50 60 70 80 90 100

Time (s)

498

499

50

501

Fre

qu

en

cy (

Hz)

CLFC

ELFC

(b) Frequency response

Fig 7 Setup and results for controller and power hardware-in-the-loop implementation

is between 1minus 45s The analysis presents an estimation of delay butdoes not provide a mean value for the delay nor a standard deviationThe variation in delay is mostly associated to the number of hopsa message packet makes in the neighborhood area network beforeit reaches the device As a constant delay is highly unlikely in areal-world scenario for this work two types of distributions are con-sidered as shown in Fig 6a a uniform distribution and a Gaussiandistribution where the mean value micro has been chosen as 275s (aver-age of minimum and maximum latency) with a standard deviation ofσ of 03s

A controller hardware-in-the-loop implementation as presentedin Fig 6b is utilized The five-areas of the GB power system aresimulated within the digital real-time simulator the proposed ELFCis implemented within a Beckhoff real-time controller while a sim-ple transactive energy based DSA runs on the host PC The proposedELFC is run at a time step of 10ms with TIaux = 5s while all othercontrol parameters remain the same as in the simulation study Theresponse of the system subject to the reference imbalance is pre-sented in Fig 6c and test metrics in Table 1 As can be observedthe response of the proposed ELFC incorporating delays charac-terized for demand side applications is stable and satisfactory andthere is not much difference in response with two different delaydistributions The deployment of the control on dedicated controllerhardware enabled by rapid prototyping further demonstrates theviability of the control for real-world implementations

IET Research Journals pp 1ndash108 ccopy The Institution of Engineering and Technology 2015

53 Controller and Power Hardware-in-the-Loop

The controller and power hardware-in-the-loop setup is shown inFig 7a As can be observed two areas of the GB network arerepresented by Cell 2 and Cell 3 of the Dynamic Power SystemsLaboratory [43] while the remaining three areas are simulated withinthe digital real-time simulator To ensure stability of the setup andto be able to represent a large portion of the GB network thecurrents are scaled by a factor of k = 08times 106 A detailed descrip-tion and process of initialization and synchronization of the powerhardware-in-the-loop setup is found in [44]

The system is subject to disturbance in LFC area 2 (Cell 2 inhardware Fig 7a) a loss of generation with magnitude of 800MW(1000W in hardware) A smaller magnitude of disturbance in exper-imental evaluation compared to the reference imbalance magnitudeis due to the stability of the setup which put a limit on the sparepower capacity reserved for imbalance event As the objective ofpower hardware-in-the-loop evaluation is to demonstrate the capa-bility of the proposed control to operate under real measurementsthe magnitude of the disturbance does not impact the integrity ofthe study However it further demonstrates the applicability of theproposed solution to a wide range of disturbances The frequencymeasurements are derived from the raw voltage measurements atCell 2 and Cell 3 nodes using 3-phase voltage transformers A real-time system controller a multi-processor based system with analogto digital sampling and digital pre-filtering at 66666micros is utilizedto process the measurements As in the simulation based studiesthe measurement window to obtain RoCoF remains Tn = 002sThe parameters of the control system remain same as in the pre-vious study The system frequency response subject to generationloss at t = 10s is shown in Fig 7b As can be observed the ADDwas able to identify the imbalance evident from the restoration offrequency much faster than CLFC Therefore demonstration of thefeasibility of ADD operation with real measurements strongly sup-ports the deployment of the proposed control approach in futurepower systems

6 Conclusions

In this paper to improve the speed of response of secondary fre-quency control a novel control has been proposed The controlincorporates a fast acting balance control loop in addition to the sec-ondary control loop The balance control loop incorporates an areadistrubance determiner capable of fast autonomous and definitiveidentification of location of the imbalance The impact of fast anddefinitive identification of location of imbalance is two-fold

bull it ensures unilateral activation of reserves only within the areawhere the imbalance initiated This allows for faster response speedbull it enables use of area power imbalance observed over tie-lines ascontrol input as opposed to conventionally utilized area control errorThis avoids sympathetic activations even at faster response speeds

The proposed enhanced load frequency control further decouples thecontrol effort of fast acting resources from that of conventional gen-erators The control effort to fast acting resources is a combination ofeffort from balance control loop and secondary control loop whilethe control effort to conventional generators is from secondary con-trol loop only This allows for the exploitation of true effectiveness offast acting devices for frequency restoration The analytical assess-ments in the work undertaken provides a high degree of confidencein the aptness of the proposed approach for dynamically changingpower systems while the experimental evaluation demonstrates itsreal-world applicability

Acknowledgment

This work was supported by the European Commission under theFP7 ELECTRA IRP (grant no 609687) Any opinions findings andconclusions or recommendations expressed in this material are those

of the authors and do not necessarily reflect those of the EuropeanCommission

7 References1 lsquoNational Transmission Network Development Planrsquo (Australian Energy

Market Operator (AEMO) 2016) Available from httpwwwaemocomauElectricityNational-Electricity-Market-NEMPlanning-and-forecastingNational-Transmission-Network-Development-Plan

2 Torres LMA Lopes LAC MorAtildean TLA Espinoza CJR lsquoSelf-tuningvirtual synchronous machine A control strategy for energy storage systems to sup-port dynamic frequency controlrsquo IEEE Transactions on Energy Conversion 201429 (4) pp 833ndash840

3 Rakhshani E Remon D Cantarellas AM Rodriguez P lsquoAnalysis of deriva-tive control based virtual inertia in multi-area high-voltage direct current intercon-nected power systemsrsquo IET Generation Transmission Distribution 2016 10 (6)pp 1458ndash1469

4 Alipoor J Miura Y Ise T lsquoPower system stabilization using virtual syn-chronous generator with alternating moment of inertiarsquo IEEE Journal of Emergingand Selected Topics in Power Electronics 2015 3 (2) pp 451ndash458

5 Soni N Doolla S Chandorkar MC lsquoImprovement of transient response inmicrogrids using virtual inertiarsquo IEEE Transactions on Power Delivery 2013 28(3) pp 1830ndash1838

6 lsquoFast Frequency Response in the NEM Working Paperrsquo (Aus-tralian Energy Market Operator (AEMO) 2017) Available fromhttpswwwaemocomau-mediaFilesElectricityNEMSecurityandReliabilityReports2017FFR-Working-Paper---Finalpdf

7 lsquoSystem Operability Framework 2016rsquo (National Grid 2016) Availablefrom httpwww2nationalgridcomUKIndustry-informationFuture-of-EnergySystem-Operability-Framework

8 lsquoRoCoF - An independent analysis on the ability of Generators to ride throughRate of Change of Frequency values up to 2Hzsrsquo (DNV-KEMA for Eir-GridSONI 2013) Available from httpwwweirgridgroupcomsite-fileslibraryEirGridDNV-KEMA_Report_RoCoF_20130208final_pdf

9 lsquoInternational Review of Frequency Control Adaptationrsquo (DGA Consultingreport to Australian Energy Market Operator (AEMO) 2016) Available fromhttpswwwaemocomau-mediaFilesElectricityNEMSecurityandReliabilityReportsFPSS---International-Review-of-FrequencyControlpdf

10 lsquoPolicy P1 Load-Frequency Control and Performancersquo (in Continen-tal Europe Operation Handbook ENTSO-E 2016) Available fromhttpswwwentsoeeufileadminuser_upload_librarypublicationsentsoeOperation_HandbookPolicy_1_finalpdf

11 lsquoBalancing and Frequency Controlrsquo (North American Electric ReliabilityCorporation 2011) Available from httpwwwnerccomdocsocrsNERC20Balancing20and20Frequency20Control20040520111pdf

12 lsquoGuide to Ancillary Services in the National Electricity Mar-ketrsquo (Australian Energy Market Operator (AEMO) 2015)Available from httpswwwaemocomau-mediaFilesPDFGuide-to-Ancillary-Services-in-the-National-Electricity-Marketashx

13 Jin C Lu N Lu S Makarov YV Dougal RA lsquoA coordinating algorithm fordispatching regulation services between slow and fast power regulating resourcesrsquoIEEE Transactions on Smart Grid 2014 5 (2) pp 1043ndash1050

14 Guo Y Xie X Wang B Shi W Dong Y Mou L et al lsquoImproving agcperformance of a coal-fueled generators with mw-level bessrsquo In 2016 IEEE PowerEnergy Society Innovative Smart Grid Technologies Conference (ISGT) ( 2016pp 1ndash5

15 Xie X Guo Y Wang B Dong Y Mou L Xue F lsquoImproving agc perfor-mance of coal-fueled thermal generators using multi-mw scale bess A practicalapplicationrsquo IEEE Transactions on Smart Grid 2018 9 (3) pp 1769ndash1777

16 Zhang F Hu Z Xie X Zhang J Song Y lsquoAssessment of the effectivenessof energy storage resources in the frequency regulation of a single-area powersystemrsquo IEEE Transactions on Power Systems 2017 32 (5) pp 3373ndash3380

17 KintnerMeyer M lsquoRegulatory policy and markets for energy storage in northamericarsquo Proceedings of the IEEE 2014 102 (7) pp 1065ndash1072

18 lsquoFrequency Control by Demand Management (FCDM)rsquo (National Grid 2015)Available from httpwww2nationalgridcomWorkAreaDownloadAssetaspxid=41294

19 Short JA Infield DG Freris LL lsquoStabilization of grid frequency throughdynamic demand controlrsquo IEEE Transactions on Power Systems 2007 22 (3)pp 1284ndash1293

20 Hu J Cao J Guerrero JM Yong T Yu J lsquoImproving frequency stabilitybased on distributed control of multiple load aggregatorsrsquo IEEE Transactions onSmart Grid 2017 8 (4) pp 1553ndash1567

21 Variani MH Tomsovic K lsquoDistributed automatic generation control usingflatness-based approach for high penetration of wind generationrsquo IEEE Transac-tions on Power Systems 2013 28 (3) pp 3002ndash3009

22 Xie P Li Y Zhu L Shi D Duan X lsquoSupplementary automatic genera-tion control using controllable energy storage in electric vehicle battery swappingstationsrsquo IET Generation Transmission Distribution 2016 10 (4) pp 1107ndash1116

23 Keyhani A Chatterjee A lsquoAutomatic generation control structure for smartpower gridsrsquo IEEE Transactions on Smart Grid 2012 3 (3) pp 1310ndash1316

24 Kok K Widergren S lsquoA society of devices Integrating intelligent distributedresources with transactive energyrsquo IEEE Power and Energy Magazine 2016 14(3) pp 34ndash45

25 R LGraham Knuth DE Patashnik O lsquoConcrete Mathematics A Foun-dation for Computer Sciencersquo (Addison-Wesley Publishing Company1990) Available from httpswwwcsientuedutw~r97002tempConcrete20Mathematics202epdf

IET Research Journals pp 1ndash10ccopy The Institution of Engineering and Technology 2015 9

1 Introduction

Increasing penetration of renewable generation sources the decreas-ing population of conventional synchronous generation (CGs) andthe consequent decrease in system inertia deteriorates the systemfrequency response In a study undertaken by the Australian EnergyMarket Operator it was shown that with the forecasted change ingeneration mix (ie a rapid uptake of non-synchronous generation)the rate of change of frequency (RoCoF) levels will increase suchthat by 2022 there would be less than two seconds for primary fre-quency control (PFC) actions to arrest frequency before it declinesbelow the defined containment band [1] The two seconds periodis much shorter than the response time typically observed for CGsIt can therefore be said that the decreasing inertia in the networkimposes new requirements on existing services In recent literaturea number of works have proposed novel synthetic-inertial controlapproaches [2ndash5] to alleviate the problem of declining system iner-tia however in practice synthetic inertia is commercially unavail-able [6] Furthermore as reported in [6] virtual inertia emulationis feasible within power networks with high inertia but it becomesincreasingly challenging to obtain an appropriate transient responsein low inertia circumstances Owing to the adaptableflexible natureof power electronic interfaced devices they can yield a wider rangeof responses and enforcing only inertial response from such devicescan be limiting and can curb further developments Other systemoperators around the world have noted this common concern ofdeclining system inertia and its effect on system frequency and thechallenges brought forwards towards frequency control [7 8] Real-ising the need for faster acting service provision system operatorsare working towards better understanding of the impact and inte-gration of fast response in the entire spectrum of frequency controlmechanisms (such as inertial PFC secondary frequency control) asis evident from the pilots and plans of many international systemoperators summarized in [9]

Load frequency control (LFC) also referred to as automaticgeneration control is a secondary frequency control (SFC) pro-cess employed to maintain the frequency of interconnected power-systems such as in Continental Europe USA and Australia [10ndash12]LFC is relatively slow designed for responding only after the fre-quency has been contained by the PFC The performance of LFC canconventionally be improved by the use of fast acting gas turbines asthe regulation providing units however their availability and cost ofoperation can restrict their implementation [13] An alternative to theconventional approach is the use of flexible demand available withinthe network In [14] and [15] it has been shown that the utilization ofa utility scale battery energy storage system (BESS) in conjunctionwith CGs improves the LFC delivering performance of the CGs In astudy that characterizes the effectiveness of utilizing battery energystorage systems for LFC compared to CGs a reduction in the reg-ulation requirement is observed even when they constitute a smallportion of total capacity [16] This has also been confirmed in studiesconducted by [17] Although the performance of LFC is improved byutilizing fast acting devices the slow operation design of LFC dueto ramp rate constraints on CGs does not allow for exploitation ofthe full potential of such devices

An approach already in use in the Great Britain (GB) power sys-tem and referred to as frequency control by demand management(FCDM) [18] provides an alternative means to exploit the powerramping potential of fast acting devices This approach is basedon determining a frequency threshold for participating fast actingdevices [19] The amount of reserves activated by such means issubtracted from the LFC regulation command In [20] it was shownthat such an approach alleviates the burden on LFC and provides abetter frequency response At present only relatively large commer-cial customers participate in FCDM and are activated by means ofunder-frequency relays The dimensioning of FCDM reserves andthe thresholds for their activation are carefully determined basedon the system power frequency characteristics In the future withhigher penetration of intermittent renewable energy resources withinthe grid this reserve dimensioning and thresholds determination willprove to be even more challenging as highly variable system power

frequency characteristics are anticipated The opportunity of partic-ipating in FCDM will be extended to the large amount of flexibilityforeseen to be available within the grid (including domestic devicessuch as electric vehicles and heat pumps) Devices interfaced bypower electronics will have no difficulty in participating throughlocal measurements however some devices that inherently do notincorporate local observables may not be able to participate In otherwords the large amount of distributed domestic flexibility withinthe network will be available to participate through the traditionalLFC (as it requires communications enablement rather than localmeasurements)

In [21ndash23] alternative structures for LFC and controls for theparticipation of fast acting devices have been proposed The par-ticipation of demand side fast acting devices in LFC structureshas further been facilitated by the establishment of demand sideaggregators (DSA) that can be perceived as any other regulatingreserve-providing unit under LFC [24]

To summarize it can be said that in the past there had been noincentive to improve the speed of response of SFC With the changesexpected within the power system in near future there is a growinginterest in improving the speed of response of frequency control ser-vices including SFC Improving the response speed of SFC (i) islimited due to the ramp rate constraints on participating CGs and (ii)can deteriorate the system frequency response due to SFC interactionwith the PFC response As discussed above even with the increas-ing presence of fast acting devices within the network the speed ofresponse of SFC within the conventional architecture of LFC cannotbe improved

To address the above gaps in this paper a novel SFC is proposedThe proposed control incorporates locational information ie thelocation of the disturbance (power imbalance due to lossadditionof generation or load is considered a disturbance in this paper) toenable the temporal enhancement of LFC The proposed controlcomprises a fast acting balance control loop (BCL) in addition tothe secondary control loop of LFC The BCL comprises an areadisturbance determiner capable of fast autonomous and definitiveidentification of disturbance location The fast identification of thedisturbance location allows for unilateral activation of reserves onlywithin the area where the disturbance has occurred at increasedspeeds which mitigates the deteriorating impact of its interactionwith PFC In addition the control effort to fast acting resourcesis a combination of effort from BCL and secondary control loopwhile the control effort to slower acting resources is from secondarycontrol loop only thereby eliminating the restrictions on ramp ratesconventionally imposed The proposed control is simple and doesnot require additional forecasting or complete network awarenessthereby allowing seamless integration within present day LFC Theperformance of the proposed control is compared with that of con-ventional LFC within a reduced five-area dynamic model of theGreat Britain power system by real-time simulations corroboratedanalytically by small-signal analysis and its real world applicabilityproven by experimental evaluation within a laboratory

2 Proposed Frequency Control Approach

In this section to present the proposed approach effectively the con-ventional LFC (referred to as CLFC henceforth) is first introducedand its shortcomings highlighted This is followed by the descriptionof the proposed control approach

21 Conventional Load Frequency Control (CLFC)

Consider an interconnected power system with M control areasindexed by i = 1 2 middot middot middot M The CLFC model of the ith controlarea is presented in Fig 1 As shown the CLFC comprises a primarycontrol loop and a secondary control loop where the aim of primarycontrol is to contain the frequency deviation caused by power imbal-ance in any control area while the secondary control is responsibleto recover the frequency back to its nominal value [5] The sys-tem dynamics of the ith area can be represented by the following

IET Research Journals pp 1ndash102 ccopy The Institution of Engineering and Technology 2015

Fig 1 Conventional load frequency control (CLFC)

differential equations

(1)∆fi (t) = minus D

2Hi∆fi (t) +

1

2Hi

(∆Pmi (t) + ∆P ren

i (t)

minus∆PLi (t)minus∆P tiei (t) + Σ∆PDSA

i (t))

(2)∆Pmi (t) = minus 1

Tbi∆Pmi (t) +

1

Tbi∆Pgi(t) +

TaiTbi

∆ ˙Pgi (t)

∆ ˙Pgi (t) = minus 1

Tgi∆Pgi (t) +

1

TgiαCGi∆Pci (t)minus 1

RiT gi∆fi (t)

(3)

where

(4)∆P tiei =

2π

s

Msumj=1j 6=i

Tij(∆fi minus∆fj

)where ∆fi is the change in frequency ∆Pci denotes control effortof secondary control ∆Pmi and ∆Pgi are the deviations of gen-erator mechanical output and valve position for area i respectivelyHi Di Ri Tgi Tai and Tbi are the inertia constant load damp-ing coefficient speed droop governor and turbine time constants forarea i respectively ∆P ren

i ∆PLi ∆P tiei and ∆PDSA

i are powervariations of renewable generation loads tie-line and DSAs respec-tively which can be viewed as external disturbances to the system∆fj is the change in frequency of area j and Tij is the synchroniz-ing torque The secondary control input is a signal referred to as thearea control error (ACE) and can be defined as

(5)ACEi (t) = βi∆fi (t) + ∆P tiei (t)

where βi is the frequency bias factor Upon occurrence of an eventthat leads to a deviation in frequency and ACE beyond a set thresh-old a proportional integral (PI) control is employed to force the ACEto zero and can be represented as

(6)∆Pci (t) = minusKPiACEi (t)minus 1

TIi

intACEi (t)

whereKPi TIi are the proportional gain and integral time constantrespectively

22 Shortcomings of Conventional Load Frequency Control

Theoretically the distinction between primary and secondary fre-quency control is their control objective ie primary containsfrequency deviations and secondary restores the frequency to itsnominal value Practically a time scale of operation is also asso-ciated with the two controls ie the primary response is expectedwithin ten seconds of the observed disturbance while secondaryresponse can vary from thirty seconds to minutes This is in gen-eral regarded as the secondary control intentionally being designed

Fig 2 Enhanced load frequency control (ELFC)

for much slower operation than primary control However this asso-ciation of time with control objective is due to practical limitationsas identified below

1 Slow disturbance location identification Within CLFC althoughthe entire synchronous area perceives the disturbance as a changein frequency only the area with the disturbance is responsible torestore the frequency This decentralized operation is achieved bymeans of ACE ie assuming appropriate calculation of β the ACEdistinguishes an internal disturbance (within the area) from an exter-nal one (outside the area) A definitive identification can only beobtained when the RoCoF df

dt = 0 and hence the delay in the oper-ation of the secondary control

2 Transient ACE following a disturbance Following a disturbancethe ACE of non-disturbance areas slowly converges to zero If thespeed of response of secondary control was increased the transientnon-zero ACE would enforce unnecessary secondary participationfrom non-disturbance areas This can further be exacerbated when βis miscalculated

3 Ramp-rate constraint of CGs Large CGs constitute the major-ity of the secondary reserves The ramp rate constraint of theseparticipating generators limits the speed of response of secondarycontrol

23 Proposed Enhanced Load Frequency Control (ELFC)

To overcome the limitations of the conventional approach a novelenhanced LFC (ELFC) as illustrated in Fig 2 is proposed The con-trol incorporates a fast acting BCL in addition to the primary andsecondary control loops The key features of the proposed approachare as follows

bull The BCL comprises an area disturbance determiner (ADD) capa-ble of fast and autonomous identification of disturbance location in acompletely decentralized manner This ensures activation of reservesunilaterally only within the area that has initiated the disturbanceallowing for faster response speeds (faster reserve activations)

bull The area power imbalance observed over tie-lines (∆P tiei ) is the

control input of BCL as opposed to the conventionally utilized ACEThe non dependence of control input on frequency eliminates dete-riorating interaction with primary control and avoids unnecessaryactivations

bull The BCL employs only fast acting demand side devices throughDSAs as regulation providing reserves eliminating any restrictionon speed of response due to ramp-rate constraints Therefore thepower regulation command of the DSAs comprises two parts (i)regulation command from secondary control loop with predefinedparticipation factor αDSAi and (ii) regulation command from theBCL

The fast acting nature of the BCL enables active contribution toobjectives of both primary and secondary control hence justify-ing its existence as an independent loop The description of the fastacting BCL is presented in the following section

IET Research Journals pp 1ndash10ccopy The Institution of Engineering and Technology 2015 3

3 Fast Acting Balance Control Loop (BCL)

In this section the operation of the fast acting BCL is explained

31 Fast Location Identification by Area DisturbanceDeterminer (ADD)

The objective of the ADD is to autonomously quickly and unam-biguously distinguish between disturbances originating within thearea and those originating elsewhere in the synchronous region Theoperating principle of the ADD is presented below

Any sudden imbalance between generator input mechanicalpower and load will lead to a deviation in system frequency thedynamics of which can be represented by the swing equation as

∆P =2H

fnom

df

dt(7)

where ∆P is the power imbalance in the system H is the inertiaconstant of the system fnom is the nominal frequency and df

dt is theRoCoF RoCoF at a specific time instant tk can be estimated as

df

dt=ftk minus ftkTn

Tn(8)

where Tn is the length of the measuring window Within a syn-chronous area A and for area under consideration i the neighbor-hood can be defined as NHi = iAN i sube A With j isin AN ias adjacent areas coupled over tie-lines with breaker state δij thechange in tie-line power flow for area i subject to the disturbancecan be calculated as

(9)∆P tiei =

sumjisinAN i

δijPtie0i minus

sumjisinAN i

δijPtiemeasi

where P tie0i and P tie

measi are the scheduled and measured tie-linepower flows respectively As the change in electric power flow isimmediate upon occurrence of power imbalance such that ∆P tie

i ge∆P tie

Thi measuring the extremum of dfdt as df

dt

lowast for ∆P tie

i dfdtlowast6=

0 the disturbance is classified as originating within the area if

(10)

dfdt

lowast∣∣∣dfdt lowast∣∣∣+ 1

minus minusdf

dt

lowast∣∣∣dfdt lowast∣∣∣+ 1

6=

lfloor∆P tie

i∣∣∆P tiei

∣∣+ 1

rfloorminus

lfloorminus∆P tie

i∣∣∆P tiei

∣∣+ 1

rfloor

where ∆P tieThi is the threshold of change in tie-line power flow set

based on the largest expected load variation in area i It should benoted that Eq (10) is an alternative form of signum function usingfloor (b c) and absolute (| |) functions Such representation is moreappropriate in this case due to its use for evaluation of real numbersexcept zero ie for ∆P tie

i dfdtlowast6= 0 An alternative representation

for convenience of implementation in embedded platforms is withthe use of Iversons brackets [25] as minus

[dfdt

lowastlt 0]

+[dfdt

lowastgt 0]6=

minus[∆P tie

i lt 0]

+[∆P tie

i gt 0] Upon satisfying Eq (10) ie the

sign of ∆P tiei is not equal to the sign of df

dt

lowast the output of the

disturbance determiner is di = ∆P tiei indicating the disturbance to

be within the area else di = 0 indicating the disturbance is out-with the area Therefore each area of A can identify disturbancesoriginating within themselves based on local measurements in adecentralized manner Although the identification is fast there mightbe a frequency dead-band (fdb) such that for measured frequencyfmeas no frequency control measure is employed when minusfdb ltfmeas minus fnom lt fdb

32 Auxiliary Controller

The output of ADD is an input to the auxiliary controller and forwhen Eq (10) is true di should be controlled to zero Employing aPI control the control effort can be represented as

∆PBCLi (t) = minusKPauxdi (t)minus

1

TIaux

intdi (t) (11)

where Kpaux TIaux are the proportional gain and integral timeconstant of the auxiliary controller respectively Contrary to the sec-ondary control loop where the value of TI is limited by the ramprate constraints of CGs TIaux can be chosen to achieve a fasterresponse The use of PI control for BCL is not mandatory and theoutput of ADD can be an input for direct load frequency control aspresented in [26] The effective power regulation command issued tothe DSAs can be represented as

∆PDSAlowast

i (t) = ∆PBCLi (t) + αDSAi∆Pci (t) (12)

33 Low Pass Filtering

To suppress the measurement noise and to smoothen the responseof BCL a low pass filter is utilized for ∆P tie

i and dfdt The filtered

signals at time instant tk can be represented as

∆P tielowasti (tk) = (1minus a) ∆P tielowast

i (tk minus 1) + a∆P tiei (tk) (13)

df

dt

lowastlowast(tk) = (1minus a)

df

dt

lowastlowast(tk minus 1) + a

df

dt(tk) (14)

where a = thTf+th

with th as the time step of control implementa-tion and Tf as the filter time constant It is important to choose thefilter time constant such that it renders satisfactory quality of inputwhile attributing least delay

4 Test System Characterization

In this paper the CLFC is the reference implementation againstwhich the performance of the proposed ELFC will be assessed In thefollowing sub-sections the test system characteristics are detailed

Fig 3 Reference five-area reduced GB power system

IET Research Journals pp 1ndash104 ccopy The Institution of Engineering and Technology 2015

41 Test Power System

For analyzing the proposed frequency control approach in this papera reduced six-machine dynamic model of the GB power system hasbeen chosen as the test grid It should be noted that the GB gridoperates as one synchronous area However in this paper the busesof the reduced model have been grouped to represent multiple LFCareas as shown in Fig 3 each of which relates to one or moreregions identified in the GB National Electricity Transmission Sys-tem boundary map presented in the Electricity Ten Year Statementof the Transmission System Operator [27]

411 Modeling The model has been developed in RSCAD andsimulated in real-time using a digital real-time simulator from RTDSTechnologies with each area comprising at least one aggregatedgenerator and an aggregated load Each aggregated generator is mod-eled as a large synchronous machine connected to the transmissionsystem via a step-up transformer (138400kV) The rating of eachgenerator is set according to a characteristic GB 2017 load flow[27] Each generator is controlled by the widely used IEEE type1 static excitation system A gas turbine and speed governor con-trol the speed and input torque of each machine The synchronousmachine excitation system and gas turbine parameters have beenobtained from [28] while the governor speed control parameters aretuned against real recorded events as will be explained in the follow-ing sub-section The transmission lines are modeled using π-sectionswith lumped resistance capacitance and inductance parameters cal-culated from the power flow data provided by the Electricity TenYear Statement for 2017 [27] as per the methodology presented in[29] The model parameters can be found in [30] and hence havenot been repeated in this paper

412 Validation The model has been validated by means oftwo tests [30]

1 Load flow analysis This test is to evaluate the steady-state per-formance of the model The load flow simulation data of the testsystem closely matches the winter peak 2017 data [27] This includesthe generation and demand data at each region and the power flowacross the boundaries of the regions

2 Dynamic frequency response evaluation The dynamic responseof the reference power system model is benchmarked against a mul-tiple real historic frequency deviation events with data recordedby phasor measurement units located at various points of the GBpower transmission network In [30] the event chosen was the tripof the England-France high voltage direct current inter-connectoron 11th of January 2016 leading to a power loss of 900 MW Inother words the model represents the real-world GB network onthe 11th of January 2016 The total generation and demand of themodel is adjusted to match the values on the day of the event (totaldemand=5956GW) The inertia constant the governor time con-stant the droop percentage and the load reference set-point param-eters are tuned to ensure the model frequency response matches thefrequency response (pre-disturbance RoCoF and frequency nadir)obtained from the phasor measurement units

42 Control Implementation

The following controls are implemented in each of the five areas ofthe GB power system with nominal frequency fnom = 50Hz Theparameters utilized are presented below

1 Primary Control Loop The synchronous generators in each ofthe areas participate proportionally to their capacity to contain thefrequency with a droop setting of 13

2 Secondary Control Loop The typical values of KPi and TIi arein the range of 01minus 10 and 50minus 200s respectively [31] [32] Thebest (fastest) value of TI = 50s has been chosen The value of KPihas been chosen as 01 Both the synchronous generator and aggre-gated load constitute the secondary reserves with αCGi = 025 and

αDSAi = 075 while fdb is selected as 01Hz [28] [33] [11]

3 Balance Control Loop With aggregated loads participating asreserves the values of KPaux and TIaux are selected as 01 and5s respectively With no ramp-rate constraints a smaller value ofTIaux has been chosen to demonstrate the effectiveness of BCLFor the ADD the values of ∆P tie

Thi = 50MW Tn = 002s andTf = 002s are selected

4 Delays Measurement and communication delays are incorpo-rated within the secondary and balance control loop The values ofACE and ∆P tie are thus updated every 2s representing the measure-ment delay [34] while the outputs of the two control loops ∆Pci

and ∆PDSAi are delayed by 1s representing communications delay

[35]

43 Incorporation of Renewable Generation Penetration

To analyze the performance and aptness of the proposed frequencycontrol approach in a more representative power system the incor-poration of renewable generation within the test power system isnecessary In this sub-section the incorporation of the impact ofrenewable generation penetration in particular the reduction in iner-tia within the test power system is explained First a multipliermicroinertia is defined that scales the system inertia constant (H) suchthat H = microinertiaH0 where H0 is the reference system inertia Toelaborate for example a value of microinertia = 09 leads to a 10reduction in system inertia Accordingly the inertia constant andpower generation of each of the six-machines within the test powersystem is reduced by 10 As sub-cycle phenomena are not underconsideration a simpler first-order equivalent model representingnon-synchronous generation [36] in each of the areas provide theremainder of the power generation

44 Demand Side Aggregators (DSAs)

DSAs in this work correspond to energy management systems foractive residential customers andor commercial buildings (referred toas prosumers) with fast-acting responsive loads modeled as dynamicloads in simulation The response dynamics of DSAs ie theresponse delay and uncertainties are subject to accurate load mod-eling and capacity estimation that are by nature time-varying andan important field of research on its own Two types of delays con-tribute towards the response delay of DSAs the communicationdelays between the DSA and the participating prosumer load andthe response time (onoff delay) of the prosumer load upon receiptof activation command from DSA The response characteristics of aprosumer load such as the minimum time a load has to stay onoffafter activation or the minimum time before the same load can becalled on for a second activation are considered a part of DSAcapacity estimation that is assumed to be guaranteed in principleie either by regulation or penalty mechanisms enforced by DSAitself However the communications delay is network-dependentand should be considered for establishing the performance of fre-quency control mechanisms Therefore in this paper it is assumedthat the required amount of restoration reserves are procured byeach of the LFC areas while the participating loads are assumed torespond to the control command immediately with no turn onoffdelays [37] considered for this work

45 Other Characteristics

A number of occurrences of sim1GW generation loss have beenexperienced by the GB grid within the last year and therefore adisturbance magnitude of 1GW within area 2 has been selected asthe reference frequency disturbance [38] The net generation loss isemulated by a step increase in load and a net load loss by a stepdecrease

IET Research Journals pp 1ndash10ccopy The Institution of Engineering and Technology 2015 5

5 Performance Evaluation

In this section the applicability of the proposed ELFC is evaluatedby means of real-time simulations and experiments within a state-of-the-art smart grid laboratory A three step approach for validation isadopted as follows

1 Simulation and analysis The feasibility of the approach is veri-fied by means of real-time simulations The response of the systemsubject to reference disturbance is evaluated followed by a sensitiv-ity analysis with respect to system inertia and size of disturbanceThis is followed by bringing forth the advantages offered by the keyfeatures of the BCL ie the ADD and use of tie-line power flow(∆P tie

i ) as control input instead of the ACE2 Controller Hardware-in-the-Loop validation Such a validationserves two main purposes (i) the validation of the controller underrealistic delays and (ii) the prototyping of the proposed controlwithin an actual hardware controller to demonstrate its real-worldapplication3 Controller and Power Hardware-in-the-Loop validation The per-formance improvement of the proposed ELFC over CLFC has beenachieved by means of incorporation of BCL that relies on ADDfor a faster disturbance detection The disturbance detection utilizesRoCoF that is regarded as a very noisy measurement Any con-trol that relies on RoCoF should therefore be validated by means ofreal measurements enabled by controller and power hardware-in-the-loop implementation Therefore the ADD of the proposed control isvalidated in this way

In all the above cases the performance of the proposed ELFC iscompared to that of CLFC To aid the assessment two key indica-tors are defined namely (i) frequency restoration time (T rest) and(ii) relative frequency overshoot (∆fovr) To asses the T rest of thecontrollers an error margin ε = 001 is defined such that T rest isthe time interval between the initiation of the disturbance to the pointwhen |∆f |= |fmeas minus fnom|lt ε and can be represented as

T rest = T rest isin R forall t gt T rest |∆f |le ε (15)

Defining the minimum and maximum frequency after a disturbanceas fmax and fmin respectively the relative overshoot is calculatedas

∆fovr =

∣∣∣∣fmax minus fnomfnom minus fmin

∣∣∣∣ (16)

As the GB power system is a relatively strongly coupled networkie a stiff network therefore the same frequency characteristics areobserved throughout the network [39] For simplicity and clarityof the figures only the frequency of the area with a disturbance isplotted for the analysis that follows

51 Simulation and analysis

511 Response to reference disturbance The system fre-quency responses subject to a generation loss at t = 10s and a loadloss at t = 110s for CLFC and ELFC are presented in Fig 4a Therestoration time for the CLFC is 83s with no overshoot The pro-posed ELFC by means of fast and accurate detection of disturbancelocation contributes to improving frequency nadir (frequency zenithfor load loss) and leads to a faster restoration time With a restora-tion time of 352s the proposed ELFC is twice as fast but presentsan overshoot of 35 An acceptable frequency response of a sys-tem subject to a disturbance is defined by an exponentially decayingfunction H(s) = fnom plusmnAeminus

1T t referred to as the trumpet curve

set as a requirement by the European Network of Transmission Sys-tem Operators for Electricity [31] Considering the response of theconventional control as reference trumpet curve parameters are cal-culated as A = 031Hz and T = 11s As can be observed from Fig4a the response of the proposed ELFC with a 35 overshoot iswell within the bound set by the trumpet curve The overshoot of theELFC can be abated and the response further improved by means

0 50 100 150 200

Time (s)

497

498

499

50

501

502

503

Fre

qu

en

cy

(H

z)

ELFC

CLFC

Trumpet Curve

Trumpet Curve

(a) Frequency response

0 50 100 150 200-005

0

005

AD

D O

utp

ut

Pti

e0 50 100 150 200

Time (s)

0

002

004

006

To

tal

Re

gu

lati

on

Po

we

r (p

u)

10 101 102 103 104

-004

0

Other LFC

ADD Outputs

LFC 2 ADD

Output

LFC 2

Other

LFCs

(b) ADD output and total regulation power for ELFC

microStep

microInertia

701

80

09 2

Re

sto

rati

on

Tim

e (

s)

90

CLFC

08 15

100

071

06 microStep

microInertia

25

1

30

09 2

Re

sto

rati

on

Tim

e (

s)

35

ELFC

08 15

40

071

0630

40

50

60

70

80

90

100

(c) Sensitivity analysis

Fig 4 Simulation and sensitivity analysis results for performanceevaluation of proposed ELFC

of faster acting PFC where primary response is provided by bat-tery energy storage systems as opposed to CGs as is the case forsimulations undertaken

The output of ADD and the total regulation power ie the sumof the regulation command from the DSAs and the secondary con-trol loop for all the areas subject to the generation loss and loadloss is shown in Fig 4b As can be observed only the output ofLFC 2 ADD is non-zero (and equal to ∆P tie

2 ) consequent to suc-cessful disturbance identification Furthermore this identification iswithin 300ms (sim 250ms) as shown This enables the fast activationof resources through the BCL to be activated unilaterally ie only inarea with disturbance as is evident from the total regulation powershown

512 Sensitivity analysis To further evaluate the performanceof the proposed ELFC a sensitivity analysis for variation in size ofdisturbance and system inertia is undertaken The size of disturbance(P step) is varied by means of a multiplier microstep such that P step =

microstepPstep0 where P step

0 is the reference disturbance magnitudeIn a similar manner a multiplier microinertia scales the system inertiaconstant (H) such that H = microinertiaH0 where H0 = 468s Theresults of the analysis for CLFC and ELFC are presented in Fig 4cAs can be observed the restoration time for CLFC increases with

IET Research Journals pp 1ndash106 ccopy The Institution of Engineering and Technology 2015

(a) CLFC with auxiliary controller-002 0 002 006

-100

-50

0

50

100

Imag

inary

Axis

(seco

nd

s-1

)

(seconds-1)

(seconds

-1)

-003 -001 0 001

-100

-50

0

50

100

Imag

inary

Axis

(seco

nd

s-1

)

(seconds-1)

(seconds

-1)

CLFC ELFC

Real Axis (seconds )-1Real Axis (seconds )-1

TIaux

=50 s

TIaux

=5 s

TIaux

=50 s

TIaux

=5 s TIaux

= 50 s rarr 5 s

TIaux

= 50 s rarr 5 s

(b) Eigenvalue analysis

0 10 20 30 40 50 60 70 80 90 100

Time (s)

497

498

499

50

501

502

Fre

qu

en

cy (

Hz)

Accurate Bias Factor

5-

Error in Bias Factor

5+

Error in Bias Factor

ELFC

Responses

CLFC

Responses

(c) Frequency response

0 10 20 30 40 50 60 70 80 90 100

Time (s)

0

002

004

006

008

01

Reg

ula

tio

n P

ow

er

(pu

)

5-

Error in Bias Factor

5+

Error in Bias Factor

Responses for

ELFC using ACE

Responses for

ELFC using ∆ Ptie

(d) Regulation power

Fig 5 Results for performance evaluation of proposed ELFC

increasing disturbance size but the decreasing system inertia doesnot have a significant impact On the other hand the performanceof proposed ELFC is significantly different the restoration time ofwhich is not affected by the increase in disturbance size while atthe same time a decrease in system inertia leads to faster restora-tion times Therefore the proposed ELFC is more robust to varyingdisturbance sizes and system characteristics a necessity for futurepower systems

With variations in disturbance size and system inertia the relativeovershoot for the proposed ELFC is always under 40 and withinthe bounds set by the respective trumpet curve While the CLFCexhibits no overshooting characteristic with decreasing system iner-tia the frequency nadir significantly decreases The proposed ELFCimproves the frequency nadir for all the cases assessed

513 Importance of ADD The objective of this sub-section isto demonstrate that the improvement in the performance of the pro-posed control is not only due to the incorporation of additional fasterauxiliary controller but the fact that the identification of the locationof the disturbance does play a very important role Therefore theimportance of ADD is demonstrated by means of conducting small-signal stability analysis on the test power system The two cases forcomparison are (i) the proposed ELFC control and (ii) the CLFCincorporating the auxiliary controller only as shown in Fig 5a Thestate-space model of M-area interconnected power system with BCL(as shown in Fig 2) can be represented as [40]

x = Ax+Bu+ Fw

y = Cx(17)

with x = [x1 x2 x3 x4 x5]T

u = [u1 u2]T

w = [∆PL1 ∆PLM

]T(18)

where x is the state vector u is the control vector andw is the disturbance vector The internal states can be rep-resented as x1 = [∆f1 ∆fM ] x2 = [∆Pm1 ∆PmM ]x3 =

[∆P tie

1 ∆P tieM

] x4 =

[intACE1

intACEM

]and

x5 =[intd1

intdM] The internal inputs can be represented

as u1 = [∆Pc1 ∆PcM ] and u2 =[∆PDSA

1 ∆PDSAM

]

The coefficient matrices A B and C can be found in [40] Thesystem is linearized around fnom in steady-state The input to thesystem is the reference disturbance in area 2 and the output is thesystem frequency

The Eigenvalues for the two cases are obtained by varying theTIaux from 50s to 20s in steps of 10s and then to 5s in steps of 5sIn Fig 5b the pole-zero map for the CLFC (left) and the proposedELFC (right) have been presented Only the states that are impactedby TIaux are shown for clearer representation As can be observedfrom Fig 5b for CLFC as the value of TIaux is reduced the polesmove towards the imaginary axis For TIaux = 10s the poles crossthe imaginary axis representing an unstable system However forthe proposed ELFC where the ADD ensures unilateral reserve acti-vations only within the area where the disturbance has occurred thepoles are very slightly impacted by the reduction in TIaux whichtherefore does not deteriorate the stability of the system

514 Choosing ∆P tie over ACE This sub-section presents ajustification for the use of ∆P tie within BCL as opposed to the con-ventionally used ACE Consider the frequency responses presentedin Fig 5c with frequency bias factor miscalculation of 5 for boththe CLFC and proposed ELFC As can be observed a 5 miscalcu-lation of β results in a plusmn17s variation in T set for the CLFC whilehas no significant impact within the proposed ELFC β is a combina-tion of the droop set for the area and the damping offered by the loadwithin the area and it is not uncommon for errors in its calculationLarger generating units within the network are bound by the gridcode to provide a set percentage response to a variation in frequencyHowever if the units do not abide by the set droop this will lead tomiscalculation of β Furthermore presently it is normal practice todetermine β on a yearly basis [11] In future systems it is expectedthat the characteristics of the network (eg system inertia) will varyvastly even within a single day It is also expected that the differencebetween the peak load and the base load will increase significantly[7] thereby resulting in the load offering a different damping effectat different times Therefore the proposed ELFC utilizing ∆P tie ismore resilient to changes expected within the future power system

IET Research Journals pp 1ndash10ccopy The Institution of Engineering and Technology 2015 7

1 15 2 25 3 35 4 45

Delay (s)

0

1000

2000

3000

4000

5000

Sam

ple

s

Uniform Distribution

Gaussian Distribution

(a) Delay distributions under consideration

(b) Controller hardware-in-the-loop setup

0 10 20 30 40 50 60 70 80 90 100

Time (s)

4975

498

4985

499

4995

50

5005

501

Fre

qu

en

cy (

Hz)

No delay

Gaussian Distribution Delay

Uniform Distribution Delay

(c) Frequency response with and without delays

Fig 6 Setup and results for controller hardware-in-the-loop imple-mentation

Furthermore if ACE was instead utilized within the BCL of theproposed scheme although the activations of reserves are guaran-teed to be unilateral within the area where the disturbance has beendetected the miscalculation of bias factor would lead to nuisanceactivations as shown in Fig 5d It should be noted that utilizing∆P tie avoids any nuisance activations

52 Controller Hardware-in-the-Loop Validation

A detailed study on experimental validation of ancillary serviceprovision capability of DSAs controlling a portfolio of highly dis-tributed demand side devices has been undertaken in [41] Thestudy identifies that validation of any control relying on DSAs forprovision of time critical ancillary services should incorporate real-istic communication delays to demonstrate robustness The resultspresented in previous sub-sections assume a fixed communicationsdelay however the communication delay referred to in this paperis the time it takes for a control command from the aggregator toreach the participating device and is highly dependent upon the com-munications architecture employed by the potential demand sideaggregator

The performance of the communications networks that connect acontrol centre to the end devices for demand side applications hasbeen analyzed in [42] and the results show that the latency expected

Table 1 Test Metrics

Delay TRest (s) ∆fovr ()

No delay 2416 287Gaussian distribution delay 4237 44Uniform distribution delay 423 436

(a) controller and power hardware-in-the-loop setup

0 10 20 30 40 50 60 70 80 90 100

Time (s)

498

499

50

501

Fre

qu

en

cy (

Hz)

CLFC

ELFC

(b) Frequency response

Fig 7 Setup and results for controller and power hardware-in-the-loop implementation

is between 1minus 45s The analysis presents an estimation of delay butdoes not provide a mean value for the delay nor a standard deviationThe variation in delay is mostly associated to the number of hopsa message packet makes in the neighborhood area network beforeit reaches the device As a constant delay is highly unlikely in areal-world scenario for this work two types of distributions are con-sidered as shown in Fig 6a a uniform distribution and a Gaussiandistribution where the mean value micro has been chosen as 275s (aver-age of minimum and maximum latency) with a standard deviation ofσ of 03s

A controller hardware-in-the-loop implementation as presentedin Fig 6b is utilized The five-areas of the GB power system aresimulated within the digital real-time simulator the proposed ELFCis implemented within a Beckhoff real-time controller while a sim-ple transactive energy based DSA runs on the host PC The proposedELFC is run at a time step of 10ms with TIaux = 5s while all othercontrol parameters remain the same as in the simulation study Theresponse of the system subject to the reference imbalance is pre-sented in Fig 6c and test metrics in Table 1 As can be observedthe response of the proposed ELFC incorporating delays charac-terized for demand side applications is stable and satisfactory andthere is not much difference in response with two different delaydistributions The deployment of the control on dedicated controllerhardware enabled by rapid prototyping further demonstrates theviability of the control for real-world implementations

IET Research Journals pp 1ndash108 ccopy The Institution of Engineering and Technology 2015

53 Controller and Power Hardware-in-the-Loop

The controller and power hardware-in-the-loop setup is shown inFig 7a As can be observed two areas of the GB network arerepresented by Cell 2 and Cell 3 of the Dynamic Power SystemsLaboratory [43] while the remaining three areas are simulated withinthe digital real-time simulator To ensure stability of the setup andto be able to represent a large portion of the GB network thecurrents are scaled by a factor of k = 08times 106 A detailed descrip-tion and process of initialization and synchronization of the powerhardware-in-the-loop setup is found in [44]

The system is subject to disturbance in LFC area 2 (Cell 2 inhardware Fig 7a) a loss of generation with magnitude of 800MW(1000W in hardware) A smaller magnitude of disturbance in exper-imental evaluation compared to the reference imbalance magnitudeis due to the stability of the setup which put a limit on the sparepower capacity reserved for imbalance event As the objective ofpower hardware-in-the-loop evaluation is to demonstrate the capa-bility of the proposed control to operate under real measurementsthe magnitude of the disturbance does not impact the integrity ofthe study However it further demonstrates the applicability of theproposed solution to a wide range of disturbances The frequencymeasurements are derived from the raw voltage measurements atCell 2 and Cell 3 nodes using 3-phase voltage transformers A real-time system controller a multi-processor based system with analogto digital sampling and digital pre-filtering at 66666micros is utilizedto process the measurements As in the simulation based studiesthe measurement window to obtain RoCoF remains Tn = 002sThe parameters of the control system remain same as in the pre-vious study The system frequency response subject to generationloss at t = 10s is shown in Fig 7b As can be observed the ADDwas able to identify the imbalance evident from the restoration offrequency much faster than CLFC Therefore demonstration of thefeasibility of ADD operation with real measurements strongly sup-ports the deployment of the proposed control approach in futurepower systems

6 Conclusions

In this paper to improve the speed of response of secondary fre-quency control a novel control has been proposed The controlincorporates a fast acting balance control loop in addition to the sec-ondary control loop The balance control loop incorporates an areadistrubance determiner capable of fast autonomous and definitiveidentification of location of the imbalance The impact of fast anddefinitive identification of location of imbalance is two-fold

bull it ensures unilateral activation of reserves only within the areawhere the imbalance initiated This allows for faster response speedbull it enables use of area power imbalance observed over tie-lines ascontrol input as opposed to conventionally utilized area control errorThis avoids sympathetic activations even at faster response speeds

The proposed enhanced load frequency control further decouples thecontrol effort of fast acting resources from that of conventional gen-erators The control effort to fast acting resources is a combination ofeffort from balance control loop and secondary control loop whilethe control effort to conventional generators is from secondary con-trol loop only This allows for the exploitation of true effectiveness offast acting devices for frequency restoration The analytical assess-ments in the work undertaken provides a high degree of confidencein the aptness of the proposed approach for dynamically changingpower systems while the experimental evaluation demonstrates itsreal-world applicability

Acknowledgment

This work was supported by the European Commission under theFP7 ELECTRA IRP (grant no 609687) Any opinions findings andconclusions or recommendations expressed in this material are those

of the authors and do not necessarily reflect those of the EuropeanCommission

7 References1 lsquoNational Transmission Network Development Planrsquo (Australian Energy

Market Operator (AEMO) 2016) Available from httpwwwaemocomauElectricityNational-Electricity-Market-NEMPlanning-and-forecastingNational-Transmission-Network-Development-Plan

2 Torres LMA Lopes LAC MorAtildean TLA Espinoza CJR lsquoSelf-tuningvirtual synchronous machine A control strategy for energy storage systems to sup-port dynamic frequency controlrsquo IEEE Transactions on Energy Conversion 201429 (4) pp 833ndash840

3 Rakhshani E Remon D Cantarellas AM Rodriguez P lsquoAnalysis of deriva-tive control based virtual inertia in multi-area high-voltage direct current intercon-nected power systemsrsquo IET Generation Transmission Distribution 2016 10 (6)pp 1458ndash1469

4 Alipoor J Miura Y Ise T lsquoPower system stabilization using virtual syn-chronous generator with alternating moment of inertiarsquo IEEE Journal of Emergingand Selected Topics in Power Electronics 2015 3 (2) pp 451ndash458

5 Soni N Doolla S Chandorkar MC lsquoImprovement of transient response inmicrogrids using virtual inertiarsquo IEEE Transactions on Power Delivery 2013 28(3) pp 1830ndash1838

6 lsquoFast Frequency Response in the NEM Working Paperrsquo (Aus-tralian Energy Market Operator (AEMO) 2017) Available fromhttpswwwaemocomau-mediaFilesElectricityNEMSecurityandReliabilityReports2017FFR-Working-Paper---Finalpdf

7 lsquoSystem Operability Framework 2016rsquo (National Grid 2016) Availablefrom httpwww2nationalgridcomUKIndustry-informationFuture-of-EnergySystem-Operability-Framework

8 lsquoRoCoF - An independent analysis on the ability of Generators to ride throughRate of Change of Frequency values up to 2Hzsrsquo (DNV-KEMA for Eir-GridSONI 2013) Available from httpwwweirgridgroupcomsite-fileslibraryEirGridDNV-KEMA_Report_RoCoF_20130208final_pdf

9 lsquoInternational Review of Frequency Control Adaptationrsquo (DGA Consultingreport to Australian Energy Market Operator (AEMO) 2016) Available fromhttpswwwaemocomau-mediaFilesElectricityNEMSecurityandReliabilityReportsFPSS---International-Review-of-FrequencyControlpdf

10 lsquoPolicy P1 Load-Frequency Control and Performancersquo (in Continen-tal Europe Operation Handbook ENTSO-E 2016) Available fromhttpswwwentsoeeufileadminuser_upload_librarypublicationsentsoeOperation_HandbookPolicy_1_finalpdf

11 lsquoBalancing and Frequency Controlrsquo (North American Electric ReliabilityCorporation 2011) Available from httpwwwnerccomdocsocrsNERC20Balancing20and20Frequency20Control20040520111pdf

12 lsquoGuide to Ancillary Services in the National Electricity Mar-ketrsquo (Australian Energy Market Operator (AEMO) 2015)Available from httpswwwaemocomau-mediaFilesPDFGuide-to-Ancillary-Services-in-the-National-Electricity-Marketashx

13 Jin C Lu N Lu S Makarov YV Dougal RA lsquoA coordinating algorithm fordispatching regulation services between slow and fast power regulating resourcesrsquoIEEE Transactions on Smart Grid 2014 5 (2) pp 1043ndash1050

14 Guo Y Xie X Wang B Shi W Dong Y Mou L et al lsquoImproving agcperformance of a coal-fueled generators with mw-level bessrsquo In 2016 IEEE PowerEnergy Society Innovative Smart Grid Technologies Conference (ISGT) ( 2016pp 1ndash5

15 Xie X Guo Y Wang B Dong Y Mou L Xue F lsquoImproving agc perfor-mance of coal-fueled thermal generators using multi-mw scale bess A practicalapplicationrsquo IEEE Transactions on Smart Grid 2018 9 (3) pp 1769ndash1777

16 Zhang F Hu Z Xie X Zhang J Song Y lsquoAssessment of the effectivenessof energy storage resources in the frequency regulation of a single-area powersystemrsquo IEEE Transactions on Power Systems 2017 32 (5) pp 3373ndash3380

17 KintnerMeyer M lsquoRegulatory policy and markets for energy storage in northamericarsquo Proceedings of the IEEE 2014 102 (7) pp 1065ndash1072

18 lsquoFrequency Control by Demand Management (FCDM)rsquo (National Grid 2015)Available from httpwww2nationalgridcomWorkAreaDownloadAssetaspxid=41294

19 Short JA Infield DG Freris LL lsquoStabilization of grid frequency throughdynamic demand controlrsquo IEEE Transactions on Power Systems 2007 22 (3)pp 1284ndash1293

20 Hu J Cao J Guerrero JM Yong T Yu J lsquoImproving frequency stabilitybased on distributed control of multiple load aggregatorsrsquo IEEE Transactions onSmart Grid 2017 8 (4) pp 1553ndash1567

21 Variani MH Tomsovic K lsquoDistributed automatic generation control usingflatness-based approach for high penetration of wind generationrsquo IEEE Transac-tions on Power Systems 2013 28 (3) pp 3002ndash3009

22 Xie P Li Y Zhu L Shi D Duan X lsquoSupplementary automatic genera-tion control using controllable energy storage in electric vehicle battery swappingstationsrsquo IET Generation Transmission Distribution 2016 10 (4) pp 1107ndash1116

23 Keyhani A Chatterjee A lsquoAutomatic generation control structure for smartpower gridsrsquo IEEE Transactions on Smart Grid 2012 3 (3) pp 1310ndash1316

24 Kok K Widergren S lsquoA society of devices Integrating intelligent distributedresources with transactive energyrsquo IEEE Power and Energy Magazine 2016 14(3) pp 34ndash45

25 R LGraham Knuth DE Patashnik O lsquoConcrete Mathematics A Foun-dation for Computer Sciencersquo (Addison-Wesley Publishing Company1990) Available from httpswwwcsientuedutw~r97002tempConcrete20Mathematics202epdf

IET Research Journals pp 1ndash10ccopy The Institution of Engineering and Technology 2015 9

Fig 1 Conventional load frequency control (CLFC)

differential equations

(1)∆fi (t) = minus D

2Hi∆fi (t) +

1

2Hi

(∆Pmi (t) + ∆P ren

i (t)

minus∆PLi (t)minus∆P tiei (t) + Σ∆PDSA

i (t))

(2)∆Pmi (t) = minus 1

Tbi∆Pmi (t) +

1

Tbi∆Pgi(t) +

TaiTbi

∆ ˙Pgi (t)

∆ ˙Pgi (t) = minus 1

Tgi∆Pgi (t) +

1

TgiαCGi∆Pci (t)minus 1

RiT gi∆fi (t)

(3)

where

(4)∆P tiei =

2π

s

Msumj=1j 6=i

Tij(∆fi minus∆fj

)where ∆fi is the change in frequency ∆Pci denotes control effortof secondary control ∆Pmi and ∆Pgi are the deviations of gen-erator mechanical output and valve position for area i respectivelyHi Di Ri Tgi Tai and Tbi are the inertia constant load damp-ing coefficient speed droop governor and turbine time constants forarea i respectively ∆P ren

i ∆PLi ∆P tiei and ∆PDSA

i are powervariations of renewable generation loads tie-line and DSAs respec-tively which can be viewed as external disturbances to the system∆fj is the change in frequency of area j and Tij is the synchroniz-ing torque The secondary control input is a signal referred to as thearea control error (ACE) and can be defined as

(5)ACEi (t) = βi∆fi (t) + ∆P tiei (t)

where βi is the frequency bias factor Upon occurrence of an eventthat leads to a deviation in frequency and ACE beyond a set thresh-old a proportional integral (PI) control is employed to force the ACEto zero and can be represented as

(6)∆Pci (t) = minusKPiACEi (t)minus 1

TIi

intACEi (t)

whereKPi TIi are the proportional gain and integral time constantrespectively

22 Shortcomings of Conventional Load Frequency Control

Theoretically the distinction between primary and secondary fre-quency control is their control objective ie primary containsfrequency deviations and secondary restores the frequency to itsnominal value Practically a time scale of operation is also asso-ciated with the two controls ie the primary response is expectedwithin ten seconds of the observed disturbance while secondaryresponse can vary from thirty seconds to minutes This is in gen-eral regarded as the secondary control intentionally being designed

Fig 2 Enhanced load frequency control (ELFC)

for much slower operation than primary control However this asso-ciation of time with control objective is due to practical limitationsas identified below

1 Slow disturbance location identification Within CLFC althoughthe entire synchronous area perceives the disturbance as a changein frequency only the area with the disturbance is responsible torestore the frequency This decentralized operation is achieved bymeans of ACE ie assuming appropriate calculation of β the ACEdistinguishes an internal disturbance (within the area) from an exter-nal one (outside the area) A definitive identification can only beobtained when the RoCoF df

dt = 0 and hence the delay in the oper-ation of the secondary control

2 Transient ACE following a disturbance Following a disturbancethe ACE of non-disturbance areas slowly converges to zero If thespeed of response of secondary control was increased the transientnon-zero ACE would enforce unnecessary secondary participationfrom non-disturbance areas This can further be exacerbated when βis miscalculated

3 Ramp-rate constraint of CGs Large CGs constitute the major-ity of the secondary reserves The ramp rate constraint of theseparticipating generators limits the speed of response of secondarycontrol

23 Proposed Enhanced Load Frequency Control (ELFC)

To overcome the limitations of the conventional approach a novelenhanced LFC (ELFC) as illustrated in Fig 2 is proposed The con-trol incorporates a fast acting BCL in addition to the primary andsecondary control loops The key features of the proposed approachare as follows

bull The BCL comprises an area disturbance determiner (ADD) capa-ble of fast and autonomous identification of disturbance location in acompletely decentralized manner This ensures activation of reservesunilaterally only within the area that has initiated the disturbanceallowing for faster response speeds (faster reserve activations)

bull The area power imbalance observed over tie-lines (∆P tiei ) is the

control input of BCL as opposed to the conventionally utilized ACEThe non dependence of control input on frequency eliminates dete-riorating interaction with primary control and avoids unnecessaryactivations

bull The BCL employs only fast acting demand side devices throughDSAs as regulation providing reserves eliminating any restrictionon speed of response due to ramp-rate constraints Therefore thepower regulation command of the DSAs comprises two parts (i)regulation command from secondary control loop with predefinedparticipation factor αDSAi and (ii) regulation command from theBCL

The fast acting nature of the BCL enables active contribution toobjectives of both primary and secondary control hence justify-ing its existence as an independent loop The description of the fastacting BCL is presented in the following section

IET Research Journals pp 1ndash10ccopy The Institution of Engineering and Technology 2015 3

3 Fast Acting Balance Control Loop (BCL)

In this section the operation of the fast acting BCL is explained

31 Fast Location Identification by Area DisturbanceDeterminer (ADD)

The objective of the ADD is to autonomously quickly and unam-biguously distinguish between disturbances originating within thearea and those originating elsewhere in the synchronous region Theoperating principle of the ADD is presented below

Any sudden imbalance between generator input mechanicalpower and load will lead to a deviation in system frequency thedynamics of which can be represented by the swing equation as

∆P =2H

fnom

df

dt(7)

where ∆P is the power imbalance in the system H is the inertiaconstant of the system fnom is the nominal frequency and df

dt is theRoCoF RoCoF at a specific time instant tk can be estimated as

df

dt=ftk minus ftkTn

Tn(8)

where Tn is the length of the measuring window Within a syn-chronous area A and for area under consideration i the neighbor-hood can be defined as NHi = iAN i sube A With j isin AN ias adjacent areas coupled over tie-lines with breaker state δij thechange in tie-line power flow for area i subject to the disturbancecan be calculated as

(9)∆P tiei =

sumjisinAN i

δijPtie0i minus

sumjisinAN i

δijPtiemeasi

where P tie0i and P tie

measi are the scheduled and measured tie-linepower flows respectively As the change in electric power flow isimmediate upon occurrence of power imbalance such that ∆P tie

i ge∆P tie

Thi measuring the extremum of dfdt as df

dt

lowast for ∆P tie

i dfdtlowast6=

0 the disturbance is classified as originating within the area if

(10)

dfdt

lowast∣∣∣dfdt lowast∣∣∣+ 1

minus minusdf

dt

lowast∣∣∣dfdt lowast∣∣∣+ 1

6=

lfloor∆P tie

i∣∣∆P tiei

∣∣+ 1

rfloorminus

lfloorminus∆P tie

i∣∣∆P tiei

∣∣+ 1

rfloor

where ∆P tieThi is the threshold of change in tie-line power flow set

based on the largest expected load variation in area i It should benoted that Eq (10) is an alternative form of signum function usingfloor (b c) and absolute (| |) functions Such representation is moreappropriate in this case due to its use for evaluation of real numbersexcept zero ie for ∆P tie

i dfdtlowast6= 0 An alternative representation

for convenience of implementation in embedded platforms is withthe use of Iversons brackets [25] as minus

[dfdt

lowastlt 0]

+[dfdt

lowastgt 0]6=

minus[∆P tie

i lt 0]

+[∆P tie

i gt 0] Upon satisfying Eq (10) ie the

sign of ∆P tiei is not equal to the sign of df

dt

lowast the output of the

disturbance determiner is di = ∆P tiei indicating the disturbance to

be within the area else di = 0 indicating the disturbance is out-with the area Therefore each area of A can identify disturbancesoriginating within themselves based on local measurements in adecentralized manner Although the identification is fast there mightbe a frequency dead-band (fdb) such that for measured frequencyfmeas no frequency control measure is employed when minusfdb ltfmeas minus fnom lt fdb

32 Auxiliary Controller

The output of ADD is an input to the auxiliary controller and forwhen Eq (10) is true di should be controlled to zero Employing aPI control the control effort can be represented as

∆PBCLi (t) = minusKPauxdi (t)minus

1

TIaux

intdi (t) (11)

where Kpaux TIaux are the proportional gain and integral timeconstant of the auxiliary controller respectively Contrary to the sec-ondary control loop where the value of TI is limited by the ramprate constraints of CGs TIaux can be chosen to achieve a fasterresponse The use of PI control for BCL is not mandatory and theoutput of ADD can be an input for direct load frequency control aspresented in [26] The effective power regulation command issued tothe DSAs can be represented as

∆PDSAlowast

i (t) = ∆PBCLi (t) + αDSAi∆Pci (t) (12)

33 Low Pass Filtering

To suppress the measurement noise and to smoothen the responseof BCL a low pass filter is utilized for ∆P tie

i and dfdt The filtered

signals at time instant tk can be represented as

∆P tielowasti (tk) = (1minus a) ∆P tielowast

i (tk minus 1) + a∆P tiei (tk) (13)

df

dt

lowastlowast(tk) = (1minus a)

df

dt

lowastlowast(tk minus 1) + a

df

dt(tk) (14)

where a = thTf+th

with th as the time step of control implementa-tion and Tf as the filter time constant It is important to choose thefilter time constant such that it renders satisfactory quality of inputwhile attributing least delay

4 Test System Characterization

In this paper the CLFC is the reference implementation againstwhich the performance of the proposed ELFC will be assessed In thefollowing sub-sections the test system characteristics are detailed

Fig 3 Reference five-area reduced GB power system

IET Research Journals pp 1ndash104 ccopy The Institution of Engineering and Technology 2015

41 Test Power System

For analyzing the proposed frequency control approach in this papera reduced six-machine dynamic model of the GB power system hasbeen chosen as the test grid It should be noted that the GB gridoperates as one synchronous area However in this paper the busesof the reduced model have been grouped to represent multiple LFCareas as shown in Fig 3 each of which relates to one or moreregions identified in the GB National Electricity Transmission Sys-tem boundary map presented in the Electricity Ten Year Statementof the Transmission System Operator [27]

411 Modeling The model has been developed in RSCAD andsimulated in real-time using a digital real-time simulator from RTDSTechnologies with each area comprising at least one aggregatedgenerator and an aggregated load Each aggregated generator is mod-eled as a large synchronous machine connected to the transmissionsystem via a step-up transformer (138400kV) The rating of eachgenerator is set according to a characteristic GB 2017 load flow[27] Each generator is controlled by the widely used IEEE type1 static excitation system A gas turbine and speed governor con-trol the speed and input torque of each machine The synchronousmachine excitation system and gas turbine parameters have beenobtained from [28] while the governor speed control parameters aretuned against real recorded events as will be explained in the follow-ing sub-section The transmission lines are modeled using π-sectionswith lumped resistance capacitance and inductance parameters cal-culated from the power flow data provided by the Electricity TenYear Statement for 2017 [27] as per the methodology presented in[29] The model parameters can be found in [30] and hence havenot been repeated in this paper

412 Validation The model has been validated by means oftwo tests [30]

1 Load flow analysis This test is to evaluate the steady-state per-formance of the model The load flow simulation data of the testsystem closely matches the winter peak 2017 data [27] This includesthe generation and demand data at each region and the power flowacross the boundaries of the regions

2 Dynamic frequency response evaluation The dynamic responseof the reference power system model is benchmarked against a mul-tiple real historic frequency deviation events with data recordedby phasor measurement units located at various points of the GBpower transmission network In [30] the event chosen was the tripof the England-France high voltage direct current inter-connectoron 11th of January 2016 leading to a power loss of 900 MW Inother words the model represents the real-world GB network onthe 11th of January 2016 The total generation and demand of themodel is adjusted to match the values on the day of the event (totaldemand=5956GW) The inertia constant the governor time con-stant the droop percentage and the load reference set-point param-eters are tuned to ensure the model frequency response matches thefrequency response (pre-disturbance RoCoF and frequency nadir)obtained from the phasor measurement units

42 Control Implementation

The following controls are implemented in each of the five areas ofthe GB power system with nominal frequency fnom = 50Hz Theparameters utilized are presented below

1 Primary Control Loop The synchronous generators in each ofthe areas participate proportionally to their capacity to contain thefrequency with a droop setting of 13

2 Secondary Control Loop The typical values of KPi and TIi arein the range of 01minus 10 and 50minus 200s respectively [31] [32] Thebest (fastest) value of TI = 50s has been chosen The value of KPihas been chosen as 01 Both the synchronous generator and aggre-gated load constitute the secondary reserves with αCGi = 025 and

αDSAi = 075 while fdb is selected as 01Hz [28] [33] [11]

3 Balance Control Loop With aggregated loads participating asreserves the values of KPaux and TIaux are selected as 01 and5s respectively With no ramp-rate constraints a smaller value ofTIaux has been chosen to demonstrate the effectiveness of BCLFor the ADD the values of ∆P tie

Thi = 50MW Tn = 002s andTf = 002s are selected

4 Delays Measurement and communication delays are incorpo-rated within the secondary and balance control loop The values ofACE and ∆P tie are thus updated every 2s representing the measure-ment delay [34] while the outputs of the two control loops ∆Pci

and ∆PDSAi are delayed by 1s representing communications delay

[35]

43 Incorporation of Renewable Generation Penetration

To analyze the performance and aptness of the proposed frequencycontrol approach in a more representative power system the incor-poration of renewable generation within the test power system isnecessary In this sub-section the incorporation of the impact ofrenewable generation penetration in particular the reduction in iner-tia within the test power system is explained First a multipliermicroinertia is defined that scales the system inertia constant (H) suchthat H = microinertiaH0 where H0 is the reference system inertia Toelaborate for example a value of microinertia = 09 leads to a 10reduction in system inertia Accordingly the inertia constant andpower generation of each of the six-machines within the test powersystem is reduced by 10 As sub-cycle phenomena are not underconsideration a simpler first-order equivalent model representingnon-synchronous generation [36] in each of the areas provide theremainder of the power generation

44 Demand Side Aggregators (DSAs)

DSAs in this work correspond to energy management systems foractive residential customers andor commercial buildings (referred toas prosumers) with fast-acting responsive loads modeled as dynamicloads in simulation The response dynamics of DSAs ie theresponse delay and uncertainties are subject to accurate load mod-eling and capacity estimation that are by nature time-varying andan important field of research on its own Two types of delays con-tribute towards the response delay of DSAs the communicationdelays between the DSA and the participating prosumer load andthe response time (onoff delay) of the prosumer load upon receiptof activation command from DSA The response characteristics of aprosumer load such as the minimum time a load has to stay onoffafter activation or the minimum time before the same load can becalled on for a second activation are considered a part of DSAcapacity estimation that is assumed to be guaranteed in principleie either by regulation or penalty mechanisms enforced by DSAitself However the communications delay is network-dependentand should be considered for establishing the performance of fre-quency control mechanisms Therefore in this paper it is assumedthat the required amount of restoration reserves are procured byeach of the LFC areas while the participating loads are assumed torespond to the control command immediately with no turn onoffdelays [37] considered for this work

45 Other Characteristics

A number of occurrences of sim1GW generation loss have beenexperienced by the GB grid within the last year and therefore adisturbance magnitude of 1GW within area 2 has been selected asthe reference frequency disturbance [38] The net generation loss isemulated by a step increase in load and a net load loss by a stepdecrease

IET Research Journals pp 1ndash10ccopy The Institution of Engineering and Technology 2015 5

5 Performance Evaluation

In this section the applicability of the proposed ELFC is evaluatedby means of real-time simulations and experiments within a state-of-the-art smart grid laboratory A three step approach for validation isadopted as follows

1 Simulation and analysis The feasibility of the approach is veri-fied by means of real-time simulations The response of the systemsubject to reference disturbance is evaluated followed by a sensitiv-ity analysis with respect to system inertia and size of disturbanceThis is followed by bringing forth the advantages offered by the keyfeatures of the BCL ie the ADD and use of tie-line power flow(∆P tie

i ) as control input instead of the ACE2 Controller Hardware-in-the-Loop validation Such a validationserves two main purposes (i) the validation of the controller underrealistic delays and (ii) the prototyping of the proposed controlwithin an actual hardware controller to demonstrate its real-worldapplication3 Controller and Power Hardware-in-the-Loop validation The per-formance improvement of the proposed ELFC over CLFC has beenachieved by means of incorporation of BCL that relies on ADDfor a faster disturbance detection The disturbance detection utilizesRoCoF that is regarded as a very noisy measurement Any con-trol that relies on RoCoF should therefore be validated by means ofreal measurements enabled by controller and power hardware-in-the-loop implementation Therefore the ADD of the proposed control isvalidated in this way

In all the above cases the performance of the proposed ELFC iscompared to that of CLFC To aid the assessment two key indica-tors are defined namely (i) frequency restoration time (T rest) and(ii) relative frequency overshoot (∆fovr) To asses the T rest of thecontrollers an error margin ε = 001 is defined such that T rest isthe time interval between the initiation of the disturbance to the pointwhen |∆f |= |fmeas minus fnom|lt ε and can be represented as

T rest = T rest isin R forall t gt T rest |∆f |le ε (15)

Defining the minimum and maximum frequency after a disturbanceas fmax and fmin respectively the relative overshoot is calculatedas

∆fovr =

∣∣∣∣fmax minus fnomfnom minus fmin

∣∣∣∣ (16)

As the GB power system is a relatively strongly coupled networkie a stiff network therefore the same frequency characteristics areobserved throughout the network [39] For simplicity and clarityof the figures only the frequency of the area with a disturbance isplotted for the analysis that follows

51 Simulation and analysis

511 Response to reference disturbance The system fre-quency responses subject to a generation loss at t = 10s and a loadloss at t = 110s for CLFC and ELFC are presented in Fig 4a Therestoration time for the CLFC is 83s with no overshoot The pro-posed ELFC by means of fast and accurate detection of disturbancelocation contributes to improving frequency nadir (frequency zenithfor load loss) and leads to a faster restoration time With a restora-tion time of 352s the proposed ELFC is twice as fast but presentsan overshoot of 35 An acceptable frequency response of a sys-tem subject to a disturbance is defined by an exponentially decayingfunction H(s) = fnom plusmnAeminus

1T t referred to as the trumpet curve

set as a requirement by the European Network of Transmission Sys-tem Operators for Electricity [31] Considering the response of theconventional control as reference trumpet curve parameters are cal-culated as A = 031Hz and T = 11s As can be observed from Fig4a the response of the proposed ELFC with a 35 overshoot iswell within the bound set by the trumpet curve The overshoot of theELFC can be abated and the response further improved by means

0 50 100 150 200

Time (s)

497

498

499

50

501

502

503

Fre

qu

en

cy

(H

z)

ELFC

CLFC

Trumpet Curve

Trumpet Curve

(a) Frequency response

0 50 100 150 200-005

0

005

AD

D O

utp

ut

Pti

e0 50 100 150 200

Time (s)

0

002

004

006

To

tal

Re

gu

lati

on

Po

we

r (p

u)

10 101 102 103 104

-004

0

Other LFC

ADD Outputs

LFC 2 ADD

Output

LFC 2

Other

LFCs

(b) ADD output and total regulation power for ELFC

microStep

microInertia

701

80

09 2

Re

sto

rati

on

Tim

e (

s)

90

CLFC

08 15

100

071

06 microStep

microInertia

25

1

30

09 2

Re

sto

rati

on

Tim

e (

s)

35

ELFC

08 15

40

071

0630

40

50

60

70

80

90

100

(c) Sensitivity analysis

Fig 4 Simulation and sensitivity analysis results for performanceevaluation of proposed ELFC

of faster acting PFC where primary response is provided by bat-tery energy storage systems as opposed to CGs as is the case forsimulations undertaken

The output of ADD and the total regulation power ie the sumof the regulation command from the DSAs and the secondary con-trol loop for all the areas subject to the generation loss and loadloss is shown in Fig 4b As can be observed only the output ofLFC 2 ADD is non-zero (and equal to ∆P tie

2 ) consequent to suc-cessful disturbance identification Furthermore this identification iswithin 300ms (sim 250ms) as shown This enables the fast activationof resources through the BCL to be activated unilaterally ie only inarea with disturbance as is evident from the total regulation powershown

512 Sensitivity analysis To further evaluate the performanceof the proposed ELFC a sensitivity analysis for variation in size ofdisturbance and system inertia is undertaken The size of disturbance(P step) is varied by means of a multiplier microstep such that P step =

microstepPstep0 where P step

0 is the reference disturbance magnitudeIn a similar manner a multiplier microinertia scales the system inertiaconstant (H) such that H = microinertiaH0 where H0 = 468s Theresults of the analysis for CLFC and ELFC are presented in Fig 4cAs can be observed the restoration time for CLFC increases with

IET Research Journals pp 1ndash106 ccopy The Institution of Engineering and Technology 2015

(a) CLFC with auxiliary controller-002 0 002 006

-100

-50

0

50

100

Imag

inary

Axis

(seco

nd

s-1

)

(seconds-1)

(seconds

-1)

-003 -001 0 001

-100

-50

0

50

100

Imag

inary

Axis

(seco

nd

s-1

)

(seconds-1)

(seconds

-1)

CLFC ELFC

Real Axis (seconds )-1Real Axis (seconds )-1

TIaux

=50 s

TIaux

=5 s

TIaux

=50 s

TIaux

=5 s TIaux

= 50 s rarr 5 s

TIaux

= 50 s rarr 5 s

(b) Eigenvalue analysis

0 10 20 30 40 50 60 70 80 90 100

Time (s)

497

498

499

50

501

502

Fre

qu

en

cy (

Hz)

Accurate Bias Factor

5-

Error in Bias Factor

5+

Error in Bias Factor

ELFC

Responses

CLFC

Responses

(c) Frequency response

0 10 20 30 40 50 60 70 80 90 100

Time (s)

0

002

004

006

008

01

Reg

ula

tio

n P

ow

er

(pu

)

5-

Error in Bias Factor

5+

Error in Bias Factor

Responses for

ELFC using ACE

Responses for

ELFC using ∆ Ptie

(d) Regulation power

Fig 5 Results for performance evaluation of proposed ELFC

increasing disturbance size but the decreasing system inertia doesnot have a significant impact On the other hand the performanceof proposed ELFC is significantly different the restoration time ofwhich is not affected by the increase in disturbance size while atthe same time a decrease in system inertia leads to faster restora-tion times Therefore the proposed ELFC is more robust to varyingdisturbance sizes and system characteristics a necessity for futurepower systems

With variations in disturbance size and system inertia the relativeovershoot for the proposed ELFC is always under 40 and withinthe bounds set by the respective trumpet curve While the CLFCexhibits no overshooting characteristic with decreasing system iner-tia the frequency nadir significantly decreases The proposed ELFCimproves the frequency nadir for all the cases assessed

513 Importance of ADD The objective of this sub-section isto demonstrate that the improvement in the performance of the pro-posed control is not only due to the incorporation of additional fasterauxiliary controller but the fact that the identification of the locationof the disturbance does play a very important role Therefore theimportance of ADD is demonstrated by means of conducting small-signal stability analysis on the test power system The two cases forcomparison are (i) the proposed ELFC control and (ii) the CLFCincorporating the auxiliary controller only as shown in Fig 5a Thestate-space model of M-area interconnected power system with BCL(as shown in Fig 2) can be represented as [40]

x = Ax+Bu+ Fw

y = Cx(17)

with x = [x1 x2 x3 x4 x5]T

u = [u1 u2]T

w = [∆PL1 ∆PLM

]T(18)

where x is the state vector u is the control vector andw is the disturbance vector The internal states can be rep-resented as x1 = [∆f1 ∆fM ] x2 = [∆Pm1 ∆PmM ]x3 =

[∆P tie

1 ∆P tieM

] x4 =

[intACE1

intACEM

]and

x5 =[intd1

intdM] The internal inputs can be represented

as u1 = [∆Pc1 ∆PcM ] and u2 =[∆PDSA

1 ∆PDSAM

]

The coefficient matrices A B and C can be found in [40] Thesystem is linearized around fnom in steady-state The input to thesystem is the reference disturbance in area 2 and the output is thesystem frequency

The Eigenvalues for the two cases are obtained by varying theTIaux from 50s to 20s in steps of 10s and then to 5s in steps of 5sIn Fig 5b the pole-zero map for the CLFC (left) and the proposedELFC (right) have been presented Only the states that are impactedby TIaux are shown for clearer representation As can be observedfrom Fig 5b for CLFC as the value of TIaux is reduced the polesmove towards the imaginary axis For TIaux = 10s the poles crossthe imaginary axis representing an unstable system However forthe proposed ELFC where the ADD ensures unilateral reserve acti-vations only within the area where the disturbance has occurred thepoles are very slightly impacted by the reduction in TIaux whichtherefore does not deteriorate the stability of the system

514 Choosing ∆P tie over ACE This sub-section presents ajustification for the use of ∆P tie within BCL as opposed to the con-ventionally used ACE Consider the frequency responses presentedin Fig 5c with frequency bias factor miscalculation of 5 for boththe CLFC and proposed ELFC As can be observed a 5 miscalcu-lation of β results in a plusmn17s variation in T set for the CLFC whilehas no significant impact within the proposed ELFC β is a combina-tion of the droop set for the area and the damping offered by the loadwithin the area and it is not uncommon for errors in its calculationLarger generating units within the network are bound by the gridcode to provide a set percentage response to a variation in frequencyHowever if the units do not abide by the set droop this will lead tomiscalculation of β Furthermore presently it is normal practice todetermine β on a yearly basis [11] In future systems it is expectedthat the characteristics of the network (eg system inertia) will varyvastly even within a single day It is also expected that the differencebetween the peak load and the base load will increase significantly[7] thereby resulting in the load offering a different damping effectat different times Therefore the proposed ELFC utilizing ∆P tie ismore resilient to changes expected within the future power system

IET Research Journals pp 1ndash10ccopy The Institution of Engineering and Technology 2015 7

1 15 2 25 3 35 4 45

Delay (s)

0

1000

2000

3000

4000

5000

Sam

ple

s

Uniform Distribution

Gaussian Distribution

(a) Delay distributions under consideration

(b) Controller hardware-in-the-loop setup

0 10 20 30 40 50 60 70 80 90 100

Time (s)

4975

498

4985

499

4995

50

5005

501

Fre

qu

en

cy (

Hz)

No delay

Gaussian Distribution Delay

Uniform Distribution Delay

(c) Frequency response with and without delays

Fig 6 Setup and results for controller hardware-in-the-loop imple-mentation

Furthermore if ACE was instead utilized within the BCL of theproposed scheme although the activations of reserves are guaran-teed to be unilateral within the area where the disturbance has beendetected the miscalculation of bias factor would lead to nuisanceactivations as shown in Fig 5d It should be noted that utilizing∆P tie avoids any nuisance activations

52 Controller Hardware-in-the-Loop Validation

A detailed study on experimental validation of ancillary serviceprovision capability of DSAs controlling a portfolio of highly dis-tributed demand side devices has been undertaken in [41] Thestudy identifies that validation of any control relying on DSAs forprovision of time critical ancillary services should incorporate real-istic communication delays to demonstrate robustness The resultspresented in previous sub-sections assume a fixed communicationsdelay however the communication delay referred to in this paperis the time it takes for a control command from the aggregator toreach the participating device and is highly dependent upon the com-munications architecture employed by the potential demand sideaggregator

The performance of the communications networks that connect acontrol centre to the end devices for demand side applications hasbeen analyzed in [42] and the results show that the latency expected

Table 1 Test Metrics

Delay TRest (s) ∆fovr ()

No delay 2416 287Gaussian distribution delay 4237 44Uniform distribution delay 423 436

(a) controller and power hardware-in-the-loop setup

0 10 20 30 40 50 60 70 80 90 100

Time (s)

498

499

50

501

Fre

qu

en

cy (

Hz)

CLFC

ELFC

(b) Frequency response

Fig 7 Setup and results for controller and power hardware-in-the-loop implementation

is between 1minus 45s The analysis presents an estimation of delay butdoes not provide a mean value for the delay nor a standard deviationThe variation in delay is mostly associated to the number of hopsa message packet makes in the neighborhood area network beforeit reaches the device As a constant delay is highly unlikely in areal-world scenario for this work two types of distributions are con-sidered as shown in Fig 6a a uniform distribution and a Gaussiandistribution where the mean value micro has been chosen as 275s (aver-age of minimum and maximum latency) with a standard deviation ofσ of 03s

A controller hardware-in-the-loop implementation as presentedin Fig 6b is utilized The five-areas of the GB power system aresimulated within the digital real-time simulator the proposed ELFCis implemented within a Beckhoff real-time controller while a sim-ple transactive energy based DSA runs on the host PC The proposedELFC is run at a time step of 10ms with TIaux = 5s while all othercontrol parameters remain the same as in the simulation study Theresponse of the system subject to the reference imbalance is pre-sented in Fig 6c and test metrics in Table 1 As can be observedthe response of the proposed ELFC incorporating delays charac-terized for demand side applications is stable and satisfactory andthere is not much difference in response with two different delaydistributions The deployment of the control on dedicated controllerhardware enabled by rapid prototyping further demonstrates theviability of the control for real-world implementations

IET Research Journals pp 1ndash108 ccopy The Institution of Engineering and Technology 2015

53 Controller and Power Hardware-in-the-Loop

The controller and power hardware-in-the-loop setup is shown inFig 7a As can be observed two areas of the GB network arerepresented by Cell 2 and Cell 3 of the Dynamic Power SystemsLaboratory [43] while the remaining three areas are simulated withinthe digital real-time simulator To ensure stability of the setup andto be able to represent a large portion of the GB network thecurrents are scaled by a factor of k = 08times 106 A detailed descrip-tion and process of initialization and synchronization of the powerhardware-in-the-loop setup is found in [44]

The system is subject to disturbance in LFC area 2 (Cell 2 inhardware Fig 7a) a loss of generation with magnitude of 800MW(1000W in hardware) A smaller magnitude of disturbance in exper-imental evaluation compared to the reference imbalance magnitudeis due to the stability of the setup which put a limit on the sparepower capacity reserved for imbalance event As the objective ofpower hardware-in-the-loop evaluation is to demonstrate the capa-bility of the proposed control to operate under real measurementsthe magnitude of the disturbance does not impact the integrity ofthe study However it further demonstrates the applicability of theproposed solution to a wide range of disturbances The frequencymeasurements are derived from the raw voltage measurements atCell 2 and Cell 3 nodes using 3-phase voltage transformers A real-time system controller a multi-processor based system with analogto digital sampling and digital pre-filtering at 66666micros is utilizedto process the measurements As in the simulation based studiesthe measurement window to obtain RoCoF remains Tn = 002sThe parameters of the control system remain same as in the pre-vious study The system frequency response subject to generationloss at t = 10s is shown in Fig 7b As can be observed the ADDwas able to identify the imbalance evident from the restoration offrequency much faster than CLFC Therefore demonstration of thefeasibility of ADD operation with real measurements strongly sup-ports the deployment of the proposed control approach in futurepower systems

6 Conclusions

In this paper to improve the speed of response of secondary fre-quency control a novel control has been proposed The controlincorporates a fast acting balance control loop in addition to the sec-ondary control loop The balance control loop incorporates an areadistrubance determiner capable of fast autonomous and definitiveidentification of location of the imbalance The impact of fast anddefinitive identification of location of imbalance is two-fold

bull it ensures unilateral activation of reserves only within the areawhere the imbalance initiated This allows for faster response speedbull it enables use of area power imbalance observed over tie-lines ascontrol input as opposed to conventionally utilized area control errorThis avoids sympathetic activations even at faster response speeds

The proposed enhanced load frequency control further decouples thecontrol effort of fast acting resources from that of conventional gen-erators The control effort to fast acting resources is a combination ofeffort from balance control loop and secondary control loop whilethe control effort to conventional generators is from secondary con-trol loop only This allows for the exploitation of true effectiveness offast acting devices for frequency restoration The analytical assess-ments in the work undertaken provides a high degree of confidencein the aptness of the proposed approach for dynamically changingpower systems while the experimental evaluation demonstrates itsreal-world applicability

Acknowledgment

This work was supported by the European Commission under theFP7 ELECTRA IRP (grant no 609687) Any opinions findings andconclusions or recommendations expressed in this material are those

of the authors and do not necessarily reflect those of the EuropeanCommission

7 References1 lsquoNational Transmission Network Development Planrsquo (Australian Energy

Market Operator (AEMO) 2016) Available from httpwwwaemocomauElectricityNational-Electricity-Market-NEMPlanning-and-forecastingNational-Transmission-Network-Development-Plan

2 Torres LMA Lopes LAC MorAtildean TLA Espinoza CJR lsquoSelf-tuningvirtual synchronous machine A control strategy for energy storage systems to sup-port dynamic frequency controlrsquo IEEE Transactions on Energy Conversion 201429 (4) pp 833ndash840

3 Rakhshani E Remon D Cantarellas AM Rodriguez P lsquoAnalysis of deriva-tive control based virtual inertia in multi-area high-voltage direct current intercon-nected power systemsrsquo IET Generation Transmission Distribution 2016 10 (6)pp 1458ndash1469

4 Alipoor J Miura Y Ise T lsquoPower system stabilization using virtual syn-chronous generator with alternating moment of inertiarsquo IEEE Journal of Emergingand Selected Topics in Power Electronics 2015 3 (2) pp 451ndash458

5 Soni N Doolla S Chandorkar MC lsquoImprovement of transient response inmicrogrids using virtual inertiarsquo IEEE Transactions on Power Delivery 2013 28(3) pp 1830ndash1838

6 lsquoFast Frequency Response in the NEM Working Paperrsquo (Aus-tralian Energy Market Operator (AEMO) 2017) Available fromhttpswwwaemocomau-mediaFilesElectricityNEMSecurityandReliabilityReports2017FFR-Working-Paper---Finalpdf

7 lsquoSystem Operability Framework 2016rsquo (National Grid 2016) Availablefrom httpwww2nationalgridcomUKIndustry-informationFuture-of-EnergySystem-Operability-Framework

8 lsquoRoCoF - An independent analysis on the ability of Generators to ride throughRate of Change of Frequency values up to 2Hzsrsquo (DNV-KEMA for Eir-GridSONI 2013) Available from httpwwweirgridgroupcomsite-fileslibraryEirGridDNV-KEMA_Report_RoCoF_20130208final_pdf

9 lsquoInternational Review of Frequency Control Adaptationrsquo (DGA Consultingreport to Australian Energy Market Operator (AEMO) 2016) Available fromhttpswwwaemocomau-mediaFilesElectricityNEMSecurityandReliabilityReportsFPSS---International-Review-of-FrequencyControlpdf

10 lsquoPolicy P1 Load-Frequency Control and Performancersquo (in Continen-tal Europe Operation Handbook ENTSO-E 2016) Available fromhttpswwwentsoeeufileadminuser_upload_librarypublicationsentsoeOperation_HandbookPolicy_1_finalpdf

11 lsquoBalancing and Frequency Controlrsquo (North American Electric ReliabilityCorporation 2011) Available from httpwwwnerccomdocsocrsNERC20Balancing20and20Frequency20Control20040520111pdf

12 lsquoGuide to Ancillary Services in the National Electricity Mar-ketrsquo (Australian Energy Market Operator (AEMO) 2015)Available from httpswwwaemocomau-mediaFilesPDFGuide-to-Ancillary-Services-in-the-National-Electricity-Marketashx

13 Jin C Lu N Lu S Makarov YV Dougal RA lsquoA coordinating algorithm fordispatching regulation services between slow and fast power regulating resourcesrsquoIEEE Transactions on Smart Grid 2014 5 (2) pp 1043ndash1050

14 Guo Y Xie X Wang B Shi W Dong Y Mou L et al lsquoImproving agcperformance of a coal-fueled generators with mw-level bessrsquo In 2016 IEEE PowerEnergy Society Innovative Smart Grid Technologies Conference (ISGT) ( 2016pp 1ndash5

15 Xie X Guo Y Wang B Dong Y Mou L Xue F lsquoImproving agc perfor-mance of coal-fueled thermal generators using multi-mw scale bess A practicalapplicationrsquo IEEE Transactions on Smart Grid 2018 9 (3) pp 1769ndash1777

16 Zhang F Hu Z Xie X Zhang J Song Y lsquoAssessment of the effectivenessof energy storage resources in the frequency regulation of a single-area powersystemrsquo IEEE Transactions on Power Systems 2017 32 (5) pp 3373ndash3380

17 KintnerMeyer M lsquoRegulatory policy and markets for energy storage in northamericarsquo Proceedings of the IEEE 2014 102 (7) pp 1065ndash1072

18 lsquoFrequency Control by Demand Management (FCDM)rsquo (National Grid 2015)Available from httpwww2nationalgridcomWorkAreaDownloadAssetaspxid=41294

19 Short JA Infield DG Freris LL lsquoStabilization of grid frequency throughdynamic demand controlrsquo IEEE Transactions on Power Systems 2007 22 (3)pp 1284ndash1293

20 Hu J Cao J Guerrero JM Yong T Yu J lsquoImproving frequency stabilitybased on distributed control of multiple load aggregatorsrsquo IEEE Transactions onSmart Grid 2017 8 (4) pp 1553ndash1567

21 Variani MH Tomsovic K lsquoDistributed automatic generation control usingflatness-based approach for high penetration of wind generationrsquo IEEE Transac-tions on Power Systems 2013 28 (3) pp 3002ndash3009

22 Xie P Li Y Zhu L Shi D Duan X lsquoSupplementary automatic genera-tion control using controllable energy storage in electric vehicle battery swappingstationsrsquo IET Generation Transmission Distribution 2016 10 (4) pp 1107ndash1116

23 Keyhani A Chatterjee A lsquoAutomatic generation control structure for smartpower gridsrsquo IEEE Transactions on Smart Grid 2012 3 (3) pp 1310ndash1316

24 Kok K Widergren S lsquoA society of devices Integrating intelligent distributedresources with transactive energyrsquo IEEE Power and Energy Magazine 2016 14(3) pp 34ndash45

25 R LGraham Knuth DE Patashnik O lsquoConcrete Mathematics A Foun-dation for Computer Sciencersquo (Addison-Wesley Publishing Company1990) Available from httpswwwcsientuedutw~r97002tempConcrete20Mathematics202epdf

IET Research Journals pp 1ndash10ccopy The Institution of Engineering and Technology 2015 9

3 Fast Acting Balance Control Loop (BCL)

In this section the operation of the fast acting BCL is explained

31 Fast Location Identification by Area DisturbanceDeterminer (ADD)

The objective of the ADD is to autonomously quickly and unam-biguously distinguish between disturbances originating within thearea and those originating elsewhere in the synchronous region Theoperating principle of the ADD is presented below

Any sudden imbalance between generator input mechanicalpower and load will lead to a deviation in system frequency thedynamics of which can be represented by the swing equation as

∆P =2H

fnom

df

dt(7)

where ∆P is the power imbalance in the system H is the inertiaconstant of the system fnom is the nominal frequency and df

dt is theRoCoF RoCoF at a specific time instant tk can be estimated as

df

dt=ftk minus ftkTn

Tn(8)

where Tn is the length of the measuring window Within a syn-chronous area A and for area under consideration i the neighbor-hood can be defined as NHi = iAN i sube A With j isin AN ias adjacent areas coupled over tie-lines with breaker state δij thechange in tie-line power flow for area i subject to the disturbancecan be calculated as

(9)∆P tiei =

sumjisinAN i

δijPtie0i minus

sumjisinAN i

δijPtiemeasi

where P tie0i and P tie

measi are the scheduled and measured tie-linepower flows respectively As the change in electric power flow isimmediate upon occurrence of power imbalance such that ∆P tie

i ge∆P tie

Thi measuring the extremum of dfdt as df

dt

lowast for ∆P tie

i dfdtlowast6=

0 the disturbance is classified as originating within the area if

(10)

dfdt

lowast∣∣∣dfdt lowast∣∣∣+ 1

minus minusdf

dt

lowast∣∣∣dfdt lowast∣∣∣+ 1

6=

lfloor∆P tie

i∣∣∆P tiei

∣∣+ 1

rfloorminus

lfloorminus∆P tie

i∣∣∆P tiei

∣∣+ 1

rfloor

where ∆P tieThi is the threshold of change in tie-line power flow set

based on the largest expected load variation in area i It should benoted that Eq (10) is an alternative form of signum function usingfloor (b c) and absolute (| |) functions Such representation is moreappropriate in this case due to its use for evaluation of real numbersexcept zero ie for ∆P tie

i dfdtlowast6= 0 An alternative representation

for convenience of implementation in embedded platforms is withthe use of Iversons brackets [25] as minus

[dfdt

lowastlt 0]

+[dfdt

lowastgt 0]6=

minus[∆P tie

i lt 0]

+[∆P tie

i gt 0] Upon satisfying Eq (10) ie the

sign of ∆P tiei is not equal to the sign of df

dt

lowast the output of the

disturbance determiner is di = ∆P tiei indicating the disturbance to

be within the area else di = 0 indicating the disturbance is out-with the area Therefore each area of A can identify disturbancesoriginating within themselves based on local measurements in adecentralized manner Although the identification is fast there mightbe a frequency dead-band (fdb) such that for measured frequencyfmeas no frequency control measure is employed when minusfdb ltfmeas minus fnom lt fdb

32 Auxiliary Controller

The output of ADD is an input to the auxiliary controller and forwhen Eq (10) is true di should be controlled to zero Employing aPI control the control effort can be represented as

∆PBCLi (t) = minusKPauxdi (t)minus

1

TIaux

intdi (t) (11)

where Kpaux TIaux are the proportional gain and integral timeconstant of the auxiliary controller respectively Contrary to the sec-ondary control loop where the value of TI is limited by the ramprate constraints of CGs TIaux can be chosen to achieve a fasterresponse The use of PI control for BCL is not mandatory and theoutput of ADD can be an input for direct load frequency control aspresented in [26] The effective power regulation command issued tothe DSAs can be represented as

∆PDSAlowast

i (t) = ∆PBCLi (t) + αDSAi∆Pci (t) (12)

33 Low Pass Filtering

To suppress the measurement noise and to smoothen the responseof BCL a low pass filter is utilized for ∆P tie

i and dfdt The filtered

signals at time instant tk can be represented as

∆P tielowasti (tk) = (1minus a) ∆P tielowast

i (tk minus 1) + a∆P tiei (tk) (13)

df

dt

lowastlowast(tk) = (1minus a)

df

dt

lowastlowast(tk minus 1) + a

df

dt(tk) (14)

where a = thTf+th

with th as the time step of control implementa-tion and Tf as the filter time constant It is important to choose thefilter time constant such that it renders satisfactory quality of inputwhile attributing least delay

4 Test System Characterization

In this paper the CLFC is the reference implementation againstwhich the performance of the proposed ELFC will be assessed In thefollowing sub-sections the test system characteristics are detailed

Fig 3 Reference five-area reduced GB power system

IET Research Journals pp 1ndash104 ccopy The Institution of Engineering and Technology 2015

41 Test Power System

For analyzing the proposed frequency control approach in this papera reduced six-machine dynamic model of the GB power system hasbeen chosen as the test grid It should be noted that the GB gridoperates as one synchronous area However in this paper the busesof the reduced model have been grouped to represent multiple LFCareas as shown in Fig 3 each of which relates to one or moreregions identified in the GB National Electricity Transmission Sys-tem boundary map presented in the Electricity Ten Year Statementof the Transmission System Operator [27]

411 Modeling The model has been developed in RSCAD andsimulated in real-time using a digital real-time simulator from RTDSTechnologies with each area comprising at least one aggregatedgenerator and an aggregated load Each aggregated generator is mod-eled as a large synchronous machine connected to the transmissionsystem via a step-up transformer (138400kV) The rating of eachgenerator is set according to a characteristic GB 2017 load flow[27] Each generator is controlled by the widely used IEEE type1 static excitation system A gas turbine and speed governor con-trol the speed and input torque of each machine The synchronousmachine excitation system and gas turbine parameters have beenobtained from [28] while the governor speed control parameters aretuned against real recorded events as will be explained in the follow-ing sub-section The transmission lines are modeled using π-sectionswith lumped resistance capacitance and inductance parameters cal-culated from the power flow data provided by the Electricity TenYear Statement for 2017 [27] as per the methodology presented in[29] The model parameters can be found in [30] and hence havenot been repeated in this paper

412 Validation The model has been validated by means oftwo tests [30]

1 Load flow analysis This test is to evaluate the steady-state per-formance of the model The load flow simulation data of the testsystem closely matches the winter peak 2017 data [27] This includesthe generation and demand data at each region and the power flowacross the boundaries of the regions

2 Dynamic frequency response evaluation The dynamic responseof the reference power system model is benchmarked against a mul-tiple real historic frequency deviation events with data recordedby phasor measurement units located at various points of the GBpower transmission network In [30] the event chosen was the tripof the England-France high voltage direct current inter-connectoron 11th of January 2016 leading to a power loss of 900 MW Inother words the model represents the real-world GB network onthe 11th of January 2016 The total generation and demand of themodel is adjusted to match the values on the day of the event (totaldemand=5956GW) The inertia constant the governor time con-stant the droop percentage and the load reference set-point param-eters are tuned to ensure the model frequency response matches thefrequency response (pre-disturbance RoCoF and frequency nadir)obtained from the phasor measurement units

42 Control Implementation

The following controls are implemented in each of the five areas ofthe GB power system with nominal frequency fnom = 50Hz Theparameters utilized are presented below

1 Primary Control Loop The synchronous generators in each ofthe areas participate proportionally to their capacity to contain thefrequency with a droop setting of 13

2 Secondary Control Loop The typical values of KPi and TIi arein the range of 01minus 10 and 50minus 200s respectively [31] [32] Thebest (fastest) value of TI = 50s has been chosen The value of KPihas been chosen as 01 Both the synchronous generator and aggre-gated load constitute the secondary reserves with αCGi = 025 and

αDSAi = 075 while fdb is selected as 01Hz [28] [33] [11]

3 Balance Control Loop With aggregated loads participating asreserves the values of KPaux and TIaux are selected as 01 and5s respectively With no ramp-rate constraints a smaller value ofTIaux has been chosen to demonstrate the effectiveness of BCLFor the ADD the values of ∆P tie

Thi = 50MW Tn = 002s andTf = 002s are selected

4 Delays Measurement and communication delays are incorpo-rated within the secondary and balance control loop The values ofACE and ∆P tie are thus updated every 2s representing the measure-ment delay [34] while the outputs of the two control loops ∆Pci

and ∆PDSAi are delayed by 1s representing communications delay

[35]

43 Incorporation of Renewable Generation Penetration

To analyze the performance and aptness of the proposed frequencycontrol approach in a more representative power system the incor-poration of renewable generation within the test power system isnecessary In this sub-section the incorporation of the impact ofrenewable generation penetration in particular the reduction in iner-tia within the test power system is explained First a multipliermicroinertia is defined that scales the system inertia constant (H) suchthat H = microinertiaH0 where H0 is the reference system inertia Toelaborate for example a value of microinertia = 09 leads to a 10reduction in system inertia Accordingly the inertia constant andpower generation of each of the six-machines within the test powersystem is reduced by 10 As sub-cycle phenomena are not underconsideration a simpler first-order equivalent model representingnon-synchronous generation [36] in each of the areas provide theremainder of the power generation

44 Demand Side Aggregators (DSAs)

DSAs in this work correspond to energy management systems foractive residential customers andor commercial buildings (referred toas prosumers) with fast-acting responsive loads modeled as dynamicloads in simulation The response dynamics of DSAs ie theresponse delay and uncertainties are subject to accurate load mod-eling and capacity estimation that are by nature time-varying andan important field of research on its own Two types of delays con-tribute towards the response delay of DSAs the communicationdelays between the DSA and the participating prosumer load andthe response time (onoff delay) of the prosumer load upon receiptof activation command from DSA The response characteristics of aprosumer load such as the minimum time a load has to stay onoffafter activation or the minimum time before the same load can becalled on for a second activation are considered a part of DSAcapacity estimation that is assumed to be guaranteed in principleie either by regulation or penalty mechanisms enforced by DSAitself However the communications delay is network-dependentand should be considered for establishing the performance of fre-quency control mechanisms Therefore in this paper it is assumedthat the required amount of restoration reserves are procured byeach of the LFC areas while the participating loads are assumed torespond to the control command immediately with no turn onoffdelays [37] considered for this work

45 Other Characteristics

A number of occurrences of sim1GW generation loss have beenexperienced by the GB grid within the last year and therefore adisturbance magnitude of 1GW within area 2 has been selected asthe reference frequency disturbance [38] The net generation loss isemulated by a step increase in load and a net load loss by a stepdecrease

IET Research Journals pp 1ndash10ccopy The Institution of Engineering and Technology 2015 5

5 Performance Evaluation

In this section the applicability of the proposed ELFC is evaluatedby means of real-time simulations and experiments within a state-of-the-art smart grid laboratory A three step approach for validation isadopted as follows

1 Simulation and analysis The feasibility of the approach is veri-fied by means of real-time simulations The response of the systemsubject to reference disturbance is evaluated followed by a sensitiv-ity analysis with respect to system inertia and size of disturbanceThis is followed by bringing forth the advantages offered by the keyfeatures of the BCL ie the ADD and use of tie-line power flow(∆P tie

i ) as control input instead of the ACE2 Controller Hardware-in-the-Loop validation Such a validationserves two main purposes (i) the validation of the controller underrealistic delays and (ii) the prototyping of the proposed controlwithin an actual hardware controller to demonstrate its real-worldapplication3 Controller and Power Hardware-in-the-Loop validation The per-formance improvement of the proposed ELFC over CLFC has beenachieved by means of incorporation of BCL that relies on ADDfor a faster disturbance detection The disturbance detection utilizesRoCoF that is regarded as a very noisy measurement Any con-trol that relies on RoCoF should therefore be validated by means ofreal measurements enabled by controller and power hardware-in-the-loop implementation Therefore the ADD of the proposed control isvalidated in this way

In all the above cases the performance of the proposed ELFC iscompared to that of CLFC To aid the assessment two key indica-tors are defined namely (i) frequency restoration time (T rest) and(ii) relative frequency overshoot (∆fovr) To asses the T rest of thecontrollers an error margin ε = 001 is defined such that T rest isthe time interval between the initiation of the disturbance to the pointwhen |∆f |= |fmeas minus fnom|lt ε and can be represented as

T rest = T rest isin R forall t gt T rest |∆f |le ε (15)

Defining the minimum and maximum frequency after a disturbanceas fmax and fmin respectively the relative overshoot is calculatedas

∆fovr =

∣∣∣∣fmax minus fnomfnom minus fmin

∣∣∣∣ (16)

As the GB power system is a relatively strongly coupled networkie a stiff network therefore the same frequency characteristics areobserved throughout the network [39] For simplicity and clarityof the figures only the frequency of the area with a disturbance isplotted for the analysis that follows

51 Simulation and analysis

511 Response to reference disturbance The system fre-quency responses subject to a generation loss at t = 10s and a loadloss at t = 110s for CLFC and ELFC are presented in Fig 4a Therestoration time for the CLFC is 83s with no overshoot The pro-posed ELFC by means of fast and accurate detection of disturbancelocation contributes to improving frequency nadir (frequency zenithfor load loss) and leads to a faster restoration time With a restora-tion time of 352s the proposed ELFC is twice as fast but presentsan overshoot of 35 An acceptable frequency response of a sys-tem subject to a disturbance is defined by an exponentially decayingfunction H(s) = fnom plusmnAeminus

1T t referred to as the trumpet curve

set as a requirement by the European Network of Transmission Sys-tem Operators for Electricity [31] Considering the response of theconventional control as reference trumpet curve parameters are cal-culated as A = 031Hz and T = 11s As can be observed from Fig4a the response of the proposed ELFC with a 35 overshoot iswell within the bound set by the trumpet curve The overshoot of theELFC can be abated and the response further improved by means

0 50 100 150 200

Time (s)

497

498

499

50

501

502

503

Fre

qu

en

cy

(H

z)

ELFC

CLFC

Trumpet Curve

Trumpet Curve

(a) Frequency response

0 50 100 150 200-005

0

005

AD

D O

utp

ut

Pti

e0 50 100 150 200

Time (s)

0

002

004

006

To

tal

Re

gu

lati

on

Po

we

r (p

u)

10 101 102 103 104

-004

0

Other LFC

ADD Outputs

LFC 2 ADD

Output

LFC 2

Other

LFCs

(b) ADD output and total regulation power for ELFC

microStep

microInertia

701

80

09 2

Re

sto

rati

on

Tim

e (

s)

90

CLFC

08 15

100

071

06 microStep

microInertia

25

1

30

09 2

Re

sto

rati

on

Tim

e (

s)

35

ELFC

08 15

40

071

0630

40

50

60

70

80

90

100

(c) Sensitivity analysis

Fig 4 Simulation and sensitivity analysis results for performanceevaluation of proposed ELFC

of faster acting PFC where primary response is provided by bat-tery energy storage systems as opposed to CGs as is the case forsimulations undertaken

The output of ADD and the total regulation power ie the sumof the regulation command from the DSAs and the secondary con-trol loop for all the areas subject to the generation loss and loadloss is shown in Fig 4b As can be observed only the output ofLFC 2 ADD is non-zero (and equal to ∆P tie

2 ) consequent to suc-cessful disturbance identification Furthermore this identification iswithin 300ms (sim 250ms) as shown This enables the fast activationof resources through the BCL to be activated unilaterally ie only inarea with disturbance as is evident from the total regulation powershown

512 Sensitivity analysis To further evaluate the performanceof the proposed ELFC a sensitivity analysis for variation in size ofdisturbance and system inertia is undertaken The size of disturbance(P step) is varied by means of a multiplier microstep such that P step =

microstepPstep0 where P step

0 is the reference disturbance magnitudeIn a similar manner a multiplier microinertia scales the system inertiaconstant (H) such that H = microinertiaH0 where H0 = 468s Theresults of the analysis for CLFC and ELFC are presented in Fig 4cAs can be observed the restoration time for CLFC increases with

IET Research Journals pp 1ndash106 ccopy The Institution of Engineering and Technology 2015

(a) CLFC with auxiliary controller-002 0 002 006

-100

-50

0

50

100

Imag

inary

Axis

(seco

nd

s-1

)

(seconds-1)

(seconds

-1)

-003 -001 0 001

-100

-50

0

50

100

Imag

inary

Axis

(seco

nd

s-1

)

(seconds-1)

(seconds

-1)

CLFC ELFC

Real Axis (seconds )-1Real Axis (seconds )-1

TIaux

=50 s

TIaux

=5 s

TIaux

=50 s

TIaux

=5 s TIaux

= 50 s rarr 5 s

TIaux

= 50 s rarr 5 s

(b) Eigenvalue analysis

0 10 20 30 40 50 60 70 80 90 100

Time (s)

497

498

499

50

501

502

Fre

qu

en

cy (

Hz)

Accurate Bias Factor

5-

Error in Bias Factor

5+

Error in Bias Factor

ELFC

Responses

CLFC

Responses

(c) Frequency response

0 10 20 30 40 50 60 70 80 90 100

Time (s)

0

002

004

006

008

01

Reg

ula

tio

n P

ow

er

(pu

)

5-

Error in Bias Factor

5+

Error in Bias Factor

Responses for

ELFC using ACE

Responses for

ELFC using ∆ Ptie

(d) Regulation power

Fig 5 Results for performance evaluation of proposed ELFC

increasing disturbance size but the decreasing system inertia doesnot have a significant impact On the other hand the performanceof proposed ELFC is significantly different the restoration time ofwhich is not affected by the increase in disturbance size while atthe same time a decrease in system inertia leads to faster restora-tion times Therefore the proposed ELFC is more robust to varyingdisturbance sizes and system characteristics a necessity for futurepower systems

With variations in disturbance size and system inertia the relativeovershoot for the proposed ELFC is always under 40 and withinthe bounds set by the respective trumpet curve While the CLFCexhibits no overshooting characteristic with decreasing system iner-tia the frequency nadir significantly decreases The proposed ELFCimproves the frequency nadir for all the cases assessed

513 Importance of ADD The objective of this sub-section isto demonstrate that the improvement in the performance of the pro-posed control is not only due to the incorporation of additional fasterauxiliary controller but the fact that the identification of the locationof the disturbance does play a very important role Therefore theimportance of ADD is demonstrated by means of conducting small-signal stability analysis on the test power system The two cases forcomparison are (i) the proposed ELFC control and (ii) the CLFCincorporating the auxiliary controller only as shown in Fig 5a Thestate-space model of M-area interconnected power system with BCL(as shown in Fig 2) can be represented as [40]

x = Ax+Bu+ Fw

y = Cx(17)

with x = [x1 x2 x3 x4 x5]T

u = [u1 u2]T

w = [∆PL1 ∆PLM

]T(18)

where x is the state vector u is the control vector andw is the disturbance vector The internal states can be rep-resented as x1 = [∆f1 ∆fM ] x2 = [∆Pm1 ∆PmM ]x3 =

[∆P tie

1 ∆P tieM

] x4 =

[intACE1

intACEM

]and

x5 =[intd1

intdM] The internal inputs can be represented

as u1 = [∆Pc1 ∆PcM ] and u2 =[∆PDSA

1 ∆PDSAM

]

The coefficient matrices A B and C can be found in [40] Thesystem is linearized around fnom in steady-state The input to thesystem is the reference disturbance in area 2 and the output is thesystem frequency

The Eigenvalues for the two cases are obtained by varying theTIaux from 50s to 20s in steps of 10s and then to 5s in steps of 5sIn Fig 5b the pole-zero map for the CLFC (left) and the proposedELFC (right) have been presented Only the states that are impactedby TIaux are shown for clearer representation As can be observedfrom Fig 5b for CLFC as the value of TIaux is reduced the polesmove towards the imaginary axis For TIaux = 10s the poles crossthe imaginary axis representing an unstable system However forthe proposed ELFC where the ADD ensures unilateral reserve acti-vations only within the area where the disturbance has occurred thepoles are very slightly impacted by the reduction in TIaux whichtherefore does not deteriorate the stability of the system

514 Choosing ∆P tie over ACE This sub-section presents ajustification for the use of ∆P tie within BCL as opposed to the con-ventionally used ACE Consider the frequency responses presentedin Fig 5c with frequency bias factor miscalculation of 5 for boththe CLFC and proposed ELFC As can be observed a 5 miscalcu-lation of β results in a plusmn17s variation in T set for the CLFC whilehas no significant impact within the proposed ELFC β is a combina-tion of the droop set for the area and the damping offered by the loadwithin the area and it is not uncommon for errors in its calculationLarger generating units within the network are bound by the gridcode to provide a set percentage response to a variation in frequencyHowever if the units do not abide by the set droop this will lead tomiscalculation of β Furthermore presently it is normal practice todetermine β on a yearly basis [11] In future systems it is expectedthat the characteristics of the network (eg system inertia) will varyvastly even within a single day It is also expected that the differencebetween the peak load and the base load will increase significantly[7] thereby resulting in the load offering a different damping effectat different times Therefore the proposed ELFC utilizing ∆P tie ismore resilient to changes expected within the future power system

IET Research Journals pp 1ndash10ccopy The Institution of Engineering and Technology 2015 7

1 15 2 25 3 35 4 45

Delay (s)

0

1000

2000

3000

4000

5000

Sam

ple

s

Uniform Distribution

Gaussian Distribution

(a) Delay distributions under consideration

(b) Controller hardware-in-the-loop setup

0 10 20 30 40 50 60 70 80 90 100

Time (s)

4975

498

4985

499

4995

50

5005

501

Fre

qu

en

cy (

Hz)

No delay

Gaussian Distribution Delay

Uniform Distribution Delay

(c) Frequency response with and without delays

Fig 6 Setup and results for controller hardware-in-the-loop imple-mentation

Furthermore if ACE was instead utilized within the BCL of theproposed scheme although the activations of reserves are guaran-teed to be unilateral within the area where the disturbance has beendetected the miscalculation of bias factor would lead to nuisanceactivations as shown in Fig 5d It should be noted that utilizing∆P tie avoids any nuisance activations

52 Controller Hardware-in-the-Loop Validation

A detailed study on experimental validation of ancillary serviceprovision capability of DSAs controlling a portfolio of highly dis-tributed demand side devices has been undertaken in [41] Thestudy identifies that validation of any control relying on DSAs forprovision of time critical ancillary services should incorporate real-istic communication delays to demonstrate robustness The resultspresented in previous sub-sections assume a fixed communicationsdelay however the communication delay referred to in this paperis the time it takes for a control command from the aggregator toreach the participating device and is highly dependent upon the com-munications architecture employed by the potential demand sideaggregator

The performance of the communications networks that connect acontrol centre to the end devices for demand side applications hasbeen analyzed in [42] and the results show that the latency expected

Table 1 Test Metrics

Delay TRest (s) ∆fovr ()

No delay 2416 287Gaussian distribution delay 4237 44Uniform distribution delay 423 436

(a) controller and power hardware-in-the-loop setup

0 10 20 30 40 50 60 70 80 90 100

Time (s)

498

499

50

501

Fre

qu

en

cy (

Hz)

CLFC

ELFC

(b) Frequency response

Fig 7 Setup and results for controller and power hardware-in-the-loop implementation

is between 1minus 45s The analysis presents an estimation of delay butdoes not provide a mean value for the delay nor a standard deviationThe variation in delay is mostly associated to the number of hopsa message packet makes in the neighborhood area network beforeit reaches the device As a constant delay is highly unlikely in areal-world scenario for this work two types of distributions are con-sidered as shown in Fig 6a a uniform distribution and a Gaussiandistribution where the mean value micro has been chosen as 275s (aver-age of minimum and maximum latency) with a standard deviation ofσ of 03s

A controller hardware-in-the-loop implementation as presentedin Fig 6b is utilized The five-areas of the GB power system aresimulated within the digital real-time simulator the proposed ELFCis implemented within a Beckhoff real-time controller while a sim-ple transactive energy based DSA runs on the host PC The proposedELFC is run at a time step of 10ms with TIaux = 5s while all othercontrol parameters remain the same as in the simulation study Theresponse of the system subject to the reference imbalance is pre-sented in Fig 6c and test metrics in Table 1 As can be observedthe response of the proposed ELFC incorporating delays charac-terized for demand side applications is stable and satisfactory andthere is not much difference in response with two different delaydistributions The deployment of the control on dedicated controllerhardware enabled by rapid prototyping further demonstrates theviability of the control for real-world implementations

IET Research Journals pp 1ndash108 ccopy The Institution of Engineering and Technology 2015

53 Controller and Power Hardware-in-the-Loop

The controller and power hardware-in-the-loop setup is shown inFig 7a As can be observed two areas of the GB network arerepresented by Cell 2 and Cell 3 of the Dynamic Power SystemsLaboratory [43] while the remaining three areas are simulated withinthe digital real-time simulator To ensure stability of the setup andto be able to represent a large portion of the GB network thecurrents are scaled by a factor of k = 08times 106 A detailed descrip-tion and process of initialization and synchronization of the powerhardware-in-the-loop setup is found in [44]

The system is subject to disturbance in LFC area 2 (Cell 2 inhardware Fig 7a) a loss of generation with magnitude of 800MW(1000W in hardware) A smaller magnitude of disturbance in exper-imental evaluation compared to the reference imbalance magnitudeis due to the stability of the setup which put a limit on the sparepower capacity reserved for imbalance event As the objective ofpower hardware-in-the-loop evaluation is to demonstrate the capa-bility of the proposed control to operate under real measurementsthe magnitude of the disturbance does not impact the integrity ofthe study However it further demonstrates the applicability of theproposed solution to a wide range of disturbances The frequencymeasurements are derived from the raw voltage measurements atCell 2 and Cell 3 nodes using 3-phase voltage transformers A real-time system controller a multi-processor based system with analogto digital sampling and digital pre-filtering at 66666micros is utilizedto process the measurements As in the simulation based studiesthe measurement window to obtain RoCoF remains Tn = 002sThe parameters of the control system remain same as in the pre-vious study The system frequency response subject to generationloss at t = 10s is shown in Fig 7b As can be observed the ADDwas able to identify the imbalance evident from the restoration offrequency much faster than CLFC Therefore demonstration of thefeasibility of ADD operation with real measurements strongly sup-ports the deployment of the proposed control approach in futurepower systems

6 Conclusions

In this paper to improve the speed of response of secondary fre-quency control a novel control has been proposed The controlincorporates a fast acting balance control loop in addition to the sec-ondary control loop The balance control loop incorporates an areadistrubance determiner capable of fast autonomous and definitiveidentification of location of the imbalance The impact of fast anddefinitive identification of location of imbalance is two-fold

bull it ensures unilateral activation of reserves only within the areawhere the imbalance initiated This allows for faster response speedbull it enables use of area power imbalance observed over tie-lines ascontrol input as opposed to conventionally utilized area control errorThis avoids sympathetic activations even at faster response speeds

The proposed enhanced load frequency control further decouples thecontrol effort of fast acting resources from that of conventional gen-erators The control effort to fast acting resources is a combination ofeffort from balance control loop and secondary control loop whilethe control effort to conventional generators is from secondary con-trol loop only This allows for the exploitation of true effectiveness offast acting devices for frequency restoration The analytical assess-ments in the work undertaken provides a high degree of confidencein the aptness of the proposed approach for dynamically changingpower systems while the experimental evaluation demonstrates itsreal-world applicability

Acknowledgment

This work was supported by the European Commission under theFP7 ELECTRA IRP (grant no 609687) Any opinions findings andconclusions or recommendations expressed in this material are those

of the authors and do not necessarily reflect those of the EuropeanCommission

7 References1 lsquoNational Transmission Network Development Planrsquo (Australian Energy

Market Operator (AEMO) 2016) Available from httpwwwaemocomauElectricityNational-Electricity-Market-NEMPlanning-and-forecastingNational-Transmission-Network-Development-Plan

2 Torres LMA Lopes LAC MorAtildean TLA Espinoza CJR lsquoSelf-tuningvirtual synchronous machine A control strategy for energy storage systems to sup-port dynamic frequency controlrsquo IEEE Transactions on Energy Conversion 201429 (4) pp 833ndash840

3 Rakhshani E Remon D Cantarellas AM Rodriguez P lsquoAnalysis of deriva-tive control based virtual inertia in multi-area high-voltage direct current intercon-nected power systemsrsquo IET Generation Transmission Distribution 2016 10 (6)pp 1458ndash1469

4 Alipoor J Miura Y Ise T lsquoPower system stabilization using virtual syn-chronous generator with alternating moment of inertiarsquo IEEE Journal of Emergingand Selected Topics in Power Electronics 2015 3 (2) pp 451ndash458

5 Soni N Doolla S Chandorkar MC lsquoImprovement of transient response inmicrogrids using virtual inertiarsquo IEEE Transactions on Power Delivery 2013 28(3) pp 1830ndash1838

6 lsquoFast Frequency Response in the NEM Working Paperrsquo (Aus-tralian Energy Market Operator (AEMO) 2017) Available fromhttpswwwaemocomau-mediaFilesElectricityNEMSecurityandReliabilityReports2017FFR-Working-Paper---Finalpdf

7 lsquoSystem Operability Framework 2016rsquo (National Grid 2016) Availablefrom httpwww2nationalgridcomUKIndustry-informationFuture-of-EnergySystem-Operability-Framework

8 lsquoRoCoF - An independent analysis on the ability of Generators to ride throughRate of Change of Frequency values up to 2Hzsrsquo (DNV-KEMA for Eir-GridSONI 2013) Available from httpwwweirgridgroupcomsite-fileslibraryEirGridDNV-KEMA_Report_RoCoF_20130208final_pdf

9 lsquoInternational Review of Frequency Control Adaptationrsquo (DGA Consultingreport to Australian Energy Market Operator (AEMO) 2016) Available fromhttpswwwaemocomau-mediaFilesElectricityNEMSecurityandReliabilityReportsFPSS---International-Review-of-FrequencyControlpdf

10 lsquoPolicy P1 Load-Frequency Control and Performancersquo (in Continen-tal Europe Operation Handbook ENTSO-E 2016) Available fromhttpswwwentsoeeufileadminuser_upload_librarypublicationsentsoeOperation_HandbookPolicy_1_finalpdf

11 lsquoBalancing and Frequency Controlrsquo (North American Electric ReliabilityCorporation 2011) Available from httpwwwnerccomdocsocrsNERC20Balancing20and20Frequency20Control20040520111pdf

12 lsquoGuide to Ancillary Services in the National Electricity Mar-ketrsquo (Australian Energy Market Operator (AEMO) 2015)Available from httpswwwaemocomau-mediaFilesPDFGuide-to-Ancillary-Services-in-the-National-Electricity-Marketashx

13 Jin C Lu N Lu S Makarov YV Dougal RA lsquoA coordinating algorithm fordispatching regulation services between slow and fast power regulating resourcesrsquoIEEE Transactions on Smart Grid 2014 5 (2) pp 1043ndash1050

14 Guo Y Xie X Wang B Shi W Dong Y Mou L et al lsquoImproving agcperformance of a coal-fueled generators with mw-level bessrsquo In 2016 IEEE PowerEnergy Society Innovative Smart Grid Technologies Conference (ISGT) ( 2016pp 1ndash5

15 Xie X Guo Y Wang B Dong Y Mou L Xue F lsquoImproving agc perfor-mance of coal-fueled thermal generators using multi-mw scale bess A practicalapplicationrsquo IEEE Transactions on Smart Grid 2018 9 (3) pp 1769ndash1777

16 Zhang F Hu Z Xie X Zhang J Song Y lsquoAssessment of the effectivenessof energy storage resources in the frequency regulation of a single-area powersystemrsquo IEEE Transactions on Power Systems 2017 32 (5) pp 3373ndash3380

17 KintnerMeyer M lsquoRegulatory policy and markets for energy storage in northamericarsquo Proceedings of the IEEE 2014 102 (7) pp 1065ndash1072

18 lsquoFrequency Control by Demand Management (FCDM)rsquo (National Grid 2015)Available from httpwww2nationalgridcomWorkAreaDownloadAssetaspxid=41294

19 Short JA Infield DG Freris LL lsquoStabilization of grid frequency throughdynamic demand controlrsquo IEEE Transactions on Power Systems 2007 22 (3)pp 1284ndash1293

20 Hu J Cao J Guerrero JM Yong T Yu J lsquoImproving frequency stabilitybased on distributed control of multiple load aggregatorsrsquo IEEE Transactions onSmart Grid 2017 8 (4) pp 1553ndash1567

21 Variani MH Tomsovic K lsquoDistributed automatic generation control usingflatness-based approach for high penetration of wind generationrsquo IEEE Transac-tions on Power Systems 2013 28 (3) pp 3002ndash3009

22 Xie P Li Y Zhu L Shi D Duan X lsquoSupplementary automatic genera-tion control using controllable energy storage in electric vehicle battery swappingstationsrsquo IET Generation Transmission Distribution 2016 10 (4) pp 1107ndash1116

23 Keyhani A Chatterjee A lsquoAutomatic generation control structure for smartpower gridsrsquo IEEE Transactions on Smart Grid 2012 3 (3) pp 1310ndash1316

24 Kok K Widergren S lsquoA society of devices Integrating intelligent distributedresources with transactive energyrsquo IEEE Power and Energy Magazine 2016 14(3) pp 34ndash45

25 R LGraham Knuth DE Patashnik O lsquoConcrete Mathematics A Foun-dation for Computer Sciencersquo (Addison-Wesley Publishing Company1990) Available from httpswwwcsientuedutw~r97002tempConcrete20Mathematics202epdf

IET Research Journals pp 1ndash10ccopy The Institution of Engineering and Technology 2015 9

41 Test Power System

For analyzing the proposed frequency control approach in this papera reduced six-machine dynamic model of the GB power system hasbeen chosen as the test grid It should be noted that the GB gridoperates as one synchronous area However in this paper the busesof the reduced model have been grouped to represent multiple LFCareas as shown in Fig 3 each of which relates to one or moreregions identified in the GB National Electricity Transmission Sys-tem boundary map presented in the Electricity Ten Year Statementof the Transmission System Operator [27]

411 Modeling The model has been developed in RSCAD andsimulated in real-time using a digital real-time simulator from RTDSTechnologies with each area comprising at least one aggregatedgenerator and an aggregated load Each aggregated generator is mod-eled as a large synchronous machine connected to the transmissionsystem via a step-up transformer (138400kV) The rating of eachgenerator is set according to a characteristic GB 2017 load flow[27] Each generator is controlled by the widely used IEEE type1 static excitation system A gas turbine and speed governor con-trol the speed and input torque of each machine The synchronousmachine excitation system and gas turbine parameters have beenobtained from [28] while the governor speed control parameters aretuned against real recorded events as will be explained in the follow-ing sub-section The transmission lines are modeled using π-sectionswith lumped resistance capacitance and inductance parameters cal-culated from the power flow data provided by the Electricity TenYear Statement for 2017 [27] as per the methodology presented in[29] The model parameters can be found in [30] and hence havenot been repeated in this paper

412 Validation The model has been validated by means oftwo tests [30]

1 Load flow analysis This test is to evaluate the steady-state per-formance of the model The load flow simulation data of the testsystem closely matches the winter peak 2017 data [27] This includesthe generation and demand data at each region and the power flowacross the boundaries of the regions

2 Dynamic frequency response evaluation The dynamic responseof the reference power system model is benchmarked against a mul-tiple real historic frequency deviation events with data recordedby phasor measurement units located at various points of the GBpower transmission network In [30] the event chosen was the tripof the England-France high voltage direct current inter-connectoron 11th of January 2016 leading to a power loss of 900 MW Inother words the model represents the real-world GB network onthe 11th of January 2016 The total generation and demand of themodel is adjusted to match the values on the day of the event (totaldemand=5956GW) The inertia constant the governor time con-stant the droop percentage and the load reference set-point param-eters are tuned to ensure the model frequency response matches thefrequency response (pre-disturbance RoCoF and frequency nadir)obtained from the phasor measurement units

42 Control Implementation

The following controls are implemented in each of the five areas ofthe GB power system with nominal frequency fnom = 50Hz Theparameters utilized are presented below

1 Primary Control Loop The synchronous generators in each ofthe areas participate proportionally to their capacity to contain thefrequency with a droop setting of 13

2 Secondary Control Loop The typical values of KPi and TIi arein the range of 01minus 10 and 50minus 200s respectively [31] [32] Thebest (fastest) value of TI = 50s has been chosen The value of KPihas been chosen as 01 Both the synchronous generator and aggre-gated load constitute the secondary reserves with αCGi = 025 and

αDSAi = 075 while fdb is selected as 01Hz [28] [33] [11]

3 Balance Control Loop With aggregated loads participating asreserves the values of KPaux and TIaux are selected as 01 and5s respectively With no ramp-rate constraints a smaller value ofTIaux has been chosen to demonstrate the effectiveness of BCLFor the ADD the values of ∆P tie

Thi = 50MW Tn = 002s andTf = 002s are selected

4 Delays Measurement and communication delays are incorpo-rated within the secondary and balance control loop The values ofACE and ∆P tie are thus updated every 2s representing the measure-ment delay [34] while the outputs of the two control loops ∆Pci

and ∆PDSAi are delayed by 1s representing communications delay

[35]

43 Incorporation of Renewable Generation Penetration

To analyze the performance and aptness of the proposed frequencycontrol approach in a more representative power system the incor-poration of renewable generation within the test power system isnecessary In this sub-section the incorporation of the impact ofrenewable generation penetration in particular the reduction in iner-tia within the test power system is explained First a multipliermicroinertia is defined that scales the system inertia constant (H) suchthat H = microinertiaH0 where H0 is the reference system inertia Toelaborate for example a value of microinertia = 09 leads to a 10reduction in system inertia Accordingly the inertia constant andpower generation of each of the six-machines within the test powersystem is reduced by 10 As sub-cycle phenomena are not underconsideration a simpler first-order equivalent model representingnon-synchronous generation [36] in each of the areas provide theremainder of the power generation

44 Demand Side Aggregators (DSAs)

DSAs in this work correspond to energy management systems foractive residential customers andor commercial buildings (referred toas prosumers) with fast-acting responsive loads modeled as dynamicloads in simulation The response dynamics of DSAs ie theresponse delay and uncertainties are subject to accurate load mod-eling and capacity estimation that are by nature time-varying andan important field of research on its own Two types of delays con-tribute towards the response delay of DSAs the communicationdelays between the DSA and the participating prosumer load andthe response time (onoff delay) of the prosumer load upon receiptof activation command from DSA The response characteristics of aprosumer load such as the minimum time a load has to stay onoffafter activation or the minimum time before the same load can becalled on for a second activation are considered a part of DSAcapacity estimation that is assumed to be guaranteed in principleie either by regulation or penalty mechanisms enforced by DSAitself However the communications delay is network-dependentand should be considered for establishing the performance of fre-quency control mechanisms Therefore in this paper it is assumedthat the required amount of restoration reserves are procured byeach of the LFC areas while the participating loads are assumed torespond to the control command immediately with no turn onoffdelays [37] considered for this work

45 Other Characteristics

A number of occurrences of sim1GW generation loss have beenexperienced by the GB grid within the last year and therefore adisturbance magnitude of 1GW within area 2 has been selected asthe reference frequency disturbance [38] The net generation loss isemulated by a step increase in load and a net load loss by a stepdecrease

IET Research Journals pp 1ndash10ccopy The Institution of Engineering and Technology 2015 5

5 Performance Evaluation

In this section the applicability of the proposed ELFC is evaluatedby means of real-time simulations and experiments within a state-of-the-art smart grid laboratory A three step approach for validation isadopted as follows

1 Simulation and analysis The feasibility of the approach is veri-fied by means of real-time simulations The response of the systemsubject to reference disturbance is evaluated followed by a sensitiv-ity analysis with respect to system inertia and size of disturbanceThis is followed by bringing forth the advantages offered by the keyfeatures of the BCL ie the ADD and use of tie-line power flow(∆P tie

i ) as control input instead of the ACE2 Controller Hardware-in-the-Loop validation Such a validationserves two main purposes (i) the validation of the controller underrealistic delays and (ii) the prototyping of the proposed controlwithin an actual hardware controller to demonstrate its real-worldapplication3 Controller and Power Hardware-in-the-Loop validation The per-formance improvement of the proposed ELFC over CLFC has beenachieved by means of incorporation of BCL that relies on ADDfor a faster disturbance detection The disturbance detection utilizesRoCoF that is regarded as a very noisy measurement Any con-trol that relies on RoCoF should therefore be validated by means ofreal measurements enabled by controller and power hardware-in-the-loop implementation Therefore the ADD of the proposed control isvalidated in this way

In all the above cases the performance of the proposed ELFC iscompared to that of CLFC To aid the assessment two key indica-tors are defined namely (i) frequency restoration time (T rest) and(ii) relative frequency overshoot (∆fovr) To asses the T rest of thecontrollers an error margin ε = 001 is defined such that T rest isthe time interval between the initiation of the disturbance to the pointwhen |∆f |= |fmeas minus fnom|lt ε and can be represented as

T rest = T rest isin R forall t gt T rest |∆f |le ε (15)

Defining the minimum and maximum frequency after a disturbanceas fmax and fmin respectively the relative overshoot is calculatedas

∆fovr =

∣∣∣∣fmax minus fnomfnom minus fmin

∣∣∣∣ (16)

As the GB power system is a relatively strongly coupled networkie a stiff network therefore the same frequency characteristics areobserved throughout the network [39] For simplicity and clarityof the figures only the frequency of the area with a disturbance isplotted for the analysis that follows

51 Simulation and analysis

511 Response to reference disturbance The system fre-quency responses subject to a generation loss at t = 10s and a loadloss at t = 110s for CLFC and ELFC are presented in Fig 4a Therestoration time for the CLFC is 83s with no overshoot The pro-posed ELFC by means of fast and accurate detection of disturbancelocation contributes to improving frequency nadir (frequency zenithfor load loss) and leads to a faster restoration time With a restora-tion time of 352s the proposed ELFC is twice as fast but presentsan overshoot of 35 An acceptable frequency response of a sys-tem subject to a disturbance is defined by an exponentially decayingfunction H(s) = fnom plusmnAeminus

1T t referred to as the trumpet curve

set as a requirement by the European Network of Transmission Sys-tem Operators for Electricity [31] Considering the response of theconventional control as reference trumpet curve parameters are cal-culated as A = 031Hz and T = 11s As can be observed from Fig4a the response of the proposed ELFC with a 35 overshoot iswell within the bound set by the trumpet curve The overshoot of theELFC can be abated and the response further improved by means

0 50 100 150 200

Time (s)

497

498

499

50

501

502

503

Fre

qu

en

cy

(H

z)

ELFC

CLFC

Trumpet Curve

Trumpet Curve

(a) Frequency response

0 50 100 150 200-005

0

005

AD

D O

utp

ut

Pti

e0 50 100 150 200

Time (s)

0

002

004

006

To

tal

Re

gu

lati

on

Po

we

r (p

u)

10 101 102 103 104

-004

0

Other LFC

ADD Outputs

LFC 2 ADD

Output

LFC 2

Other

LFCs

(b) ADD output and total regulation power for ELFC

microStep

microInertia

701

80

09 2

Re

sto

rati

on

Tim

e (

s)

90

CLFC

08 15

100

071

06 microStep

microInertia

25

1

30

09 2

Re

sto

rati

on

Tim

e (

s)

35

ELFC

08 15

40

071

0630

40

50

60

70

80

90

100

(c) Sensitivity analysis

Fig 4 Simulation and sensitivity analysis results for performanceevaluation of proposed ELFC

of faster acting PFC where primary response is provided by bat-tery energy storage systems as opposed to CGs as is the case forsimulations undertaken

The output of ADD and the total regulation power ie the sumof the regulation command from the DSAs and the secondary con-trol loop for all the areas subject to the generation loss and loadloss is shown in Fig 4b As can be observed only the output ofLFC 2 ADD is non-zero (and equal to ∆P tie

2 ) consequent to suc-cessful disturbance identification Furthermore this identification iswithin 300ms (sim 250ms) as shown This enables the fast activationof resources through the BCL to be activated unilaterally ie only inarea with disturbance as is evident from the total regulation powershown

512 Sensitivity analysis To further evaluate the performanceof the proposed ELFC a sensitivity analysis for variation in size ofdisturbance and system inertia is undertaken The size of disturbance(P step) is varied by means of a multiplier microstep such that P step =

microstepPstep0 where P step

0 is the reference disturbance magnitudeIn a similar manner a multiplier microinertia scales the system inertiaconstant (H) such that H = microinertiaH0 where H0 = 468s Theresults of the analysis for CLFC and ELFC are presented in Fig 4cAs can be observed the restoration time for CLFC increases with

IET Research Journals pp 1ndash106 ccopy The Institution of Engineering and Technology 2015

(a) CLFC with auxiliary controller-002 0 002 006

-100

-50

0

50

100

Imag

inary

Axis

(seco

nd

s-1

)

(seconds-1)

(seconds

-1)

-003 -001 0 001

-100

-50

0

50

100

Imag

inary

Axis

(seco

nd

s-1

)

(seconds-1)

(seconds

-1)

CLFC ELFC

Real Axis (seconds )-1Real Axis (seconds )-1

TIaux

=50 s

TIaux

=5 s

TIaux

=50 s

TIaux

=5 s TIaux

= 50 s rarr 5 s

TIaux

= 50 s rarr 5 s

(b) Eigenvalue analysis

0 10 20 30 40 50 60 70 80 90 100

Time (s)

497

498

499

50

501

502

Fre

qu

en

cy (

Hz)

Accurate Bias Factor

5-

Error in Bias Factor

5+

Error in Bias Factor

ELFC

Responses

CLFC

Responses

(c) Frequency response

0 10 20 30 40 50 60 70 80 90 100

Time (s)

0

002

004

006

008

01

Reg

ula

tio

n P

ow

er

(pu

)

5-

Error in Bias Factor

5+

Error in Bias Factor

Responses for

ELFC using ACE

Responses for

ELFC using ∆ Ptie

(d) Regulation power

Fig 5 Results for performance evaluation of proposed ELFC

increasing disturbance size but the decreasing system inertia doesnot have a significant impact On the other hand the performanceof proposed ELFC is significantly different the restoration time ofwhich is not affected by the increase in disturbance size while atthe same time a decrease in system inertia leads to faster restora-tion times Therefore the proposed ELFC is more robust to varyingdisturbance sizes and system characteristics a necessity for futurepower systems

With variations in disturbance size and system inertia the relativeovershoot for the proposed ELFC is always under 40 and withinthe bounds set by the respective trumpet curve While the CLFCexhibits no overshooting characteristic with decreasing system iner-tia the frequency nadir significantly decreases The proposed ELFCimproves the frequency nadir for all the cases assessed

513 Importance of ADD The objective of this sub-section isto demonstrate that the improvement in the performance of the pro-posed control is not only due to the incorporation of additional fasterauxiliary controller but the fact that the identification of the locationof the disturbance does play a very important role Therefore theimportance of ADD is demonstrated by means of conducting small-signal stability analysis on the test power system The two cases forcomparison are (i) the proposed ELFC control and (ii) the CLFCincorporating the auxiliary controller only as shown in Fig 5a Thestate-space model of M-area interconnected power system with BCL(as shown in Fig 2) can be represented as [40]

x = Ax+Bu+ Fw

y = Cx(17)

with x = [x1 x2 x3 x4 x5]T

u = [u1 u2]T

w = [∆PL1 ∆PLM

]T(18)

where x is the state vector u is the control vector andw is the disturbance vector The internal states can be rep-resented as x1 = [∆f1 ∆fM ] x2 = [∆Pm1 ∆PmM ]x3 =

[∆P tie

1 ∆P tieM

] x4 =

[intACE1

intACEM

]and

x5 =[intd1

intdM] The internal inputs can be represented

as u1 = [∆Pc1 ∆PcM ] and u2 =[∆PDSA

1 ∆PDSAM

]

The coefficient matrices A B and C can be found in [40] Thesystem is linearized around fnom in steady-state The input to thesystem is the reference disturbance in area 2 and the output is thesystem frequency

The Eigenvalues for the two cases are obtained by varying theTIaux from 50s to 20s in steps of 10s and then to 5s in steps of 5sIn Fig 5b the pole-zero map for the CLFC (left) and the proposedELFC (right) have been presented Only the states that are impactedby TIaux are shown for clearer representation As can be observedfrom Fig 5b for CLFC as the value of TIaux is reduced the polesmove towards the imaginary axis For TIaux = 10s the poles crossthe imaginary axis representing an unstable system However forthe proposed ELFC where the ADD ensures unilateral reserve acti-vations only within the area where the disturbance has occurred thepoles are very slightly impacted by the reduction in TIaux whichtherefore does not deteriorate the stability of the system

514 Choosing ∆P tie over ACE This sub-section presents ajustification for the use of ∆P tie within BCL as opposed to the con-ventionally used ACE Consider the frequency responses presentedin Fig 5c with frequency bias factor miscalculation of 5 for boththe CLFC and proposed ELFC As can be observed a 5 miscalcu-lation of β results in a plusmn17s variation in T set for the CLFC whilehas no significant impact within the proposed ELFC β is a combina-tion of the droop set for the area and the damping offered by the loadwithin the area and it is not uncommon for errors in its calculationLarger generating units within the network are bound by the gridcode to provide a set percentage response to a variation in frequencyHowever if the units do not abide by the set droop this will lead tomiscalculation of β Furthermore presently it is normal practice todetermine β on a yearly basis [11] In future systems it is expectedthat the characteristics of the network (eg system inertia) will varyvastly even within a single day It is also expected that the differencebetween the peak load and the base load will increase significantly[7] thereby resulting in the load offering a different damping effectat different times Therefore the proposed ELFC utilizing ∆P tie ismore resilient to changes expected within the future power system

IET Research Journals pp 1ndash10ccopy The Institution of Engineering and Technology 2015 7

1 15 2 25 3 35 4 45

Delay (s)

0

1000

2000

3000

4000

5000

Sam

ple

s

Uniform Distribution

Gaussian Distribution

(a) Delay distributions under consideration

(b) Controller hardware-in-the-loop setup

0 10 20 30 40 50 60 70 80 90 100

Time (s)

4975

498

4985

499

4995

50

5005

501

Fre

qu

en

cy (

Hz)

No delay

Gaussian Distribution Delay

Uniform Distribution Delay

(c) Frequency response with and without delays

Fig 6 Setup and results for controller hardware-in-the-loop imple-mentation

Furthermore if ACE was instead utilized within the BCL of theproposed scheme although the activations of reserves are guaran-teed to be unilateral within the area where the disturbance has beendetected the miscalculation of bias factor would lead to nuisanceactivations as shown in Fig 5d It should be noted that utilizing∆P tie avoids any nuisance activations

52 Controller Hardware-in-the-Loop Validation

A detailed study on experimental validation of ancillary serviceprovision capability of DSAs controlling a portfolio of highly dis-tributed demand side devices has been undertaken in [41] Thestudy identifies that validation of any control relying on DSAs forprovision of time critical ancillary services should incorporate real-istic communication delays to demonstrate robustness The resultspresented in previous sub-sections assume a fixed communicationsdelay however the communication delay referred to in this paperis the time it takes for a control command from the aggregator toreach the participating device and is highly dependent upon the com-munications architecture employed by the potential demand sideaggregator

The performance of the communications networks that connect acontrol centre to the end devices for demand side applications hasbeen analyzed in [42] and the results show that the latency expected

Table 1 Test Metrics

Delay TRest (s) ∆fovr ()

No delay 2416 287Gaussian distribution delay 4237 44Uniform distribution delay 423 436

(a) controller and power hardware-in-the-loop setup

0 10 20 30 40 50 60 70 80 90 100

Time (s)

498

499

50

501

Fre

qu

en

cy (

Hz)

CLFC

ELFC

(b) Frequency response

Fig 7 Setup and results for controller and power hardware-in-the-loop implementation

is between 1minus 45s The analysis presents an estimation of delay butdoes not provide a mean value for the delay nor a standard deviationThe variation in delay is mostly associated to the number of hopsa message packet makes in the neighborhood area network beforeit reaches the device As a constant delay is highly unlikely in areal-world scenario for this work two types of distributions are con-sidered as shown in Fig 6a a uniform distribution and a Gaussiandistribution where the mean value micro has been chosen as 275s (aver-age of minimum and maximum latency) with a standard deviation ofσ of 03s

A controller hardware-in-the-loop implementation as presentedin Fig 6b is utilized The five-areas of the GB power system aresimulated within the digital real-time simulator the proposed ELFCis implemented within a Beckhoff real-time controller while a sim-ple transactive energy based DSA runs on the host PC The proposedELFC is run at a time step of 10ms with TIaux = 5s while all othercontrol parameters remain the same as in the simulation study Theresponse of the system subject to the reference imbalance is pre-sented in Fig 6c and test metrics in Table 1 As can be observedthe response of the proposed ELFC incorporating delays charac-terized for demand side applications is stable and satisfactory andthere is not much difference in response with two different delaydistributions The deployment of the control on dedicated controllerhardware enabled by rapid prototyping further demonstrates theviability of the control for real-world implementations

IET Research Journals pp 1ndash108 ccopy The Institution of Engineering and Technology 2015

53 Controller and Power Hardware-in-the-Loop

The controller and power hardware-in-the-loop setup is shown inFig 7a As can be observed two areas of the GB network arerepresented by Cell 2 and Cell 3 of the Dynamic Power SystemsLaboratory [43] while the remaining three areas are simulated withinthe digital real-time simulator To ensure stability of the setup andto be able to represent a large portion of the GB network thecurrents are scaled by a factor of k = 08times 106 A detailed descrip-tion and process of initialization and synchronization of the powerhardware-in-the-loop setup is found in [44]

The system is subject to disturbance in LFC area 2 (Cell 2 inhardware Fig 7a) a loss of generation with magnitude of 800MW(1000W in hardware) A smaller magnitude of disturbance in exper-imental evaluation compared to the reference imbalance magnitudeis due to the stability of the setup which put a limit on the sparepower capacity reserved for imbalance event As the objective ofpower hardware-in-the-loop evaluation is to demonstrate the capa-bility of the proposed control to operate under real measurementsthe magnitude of the disturbance does not impact the integrity ofthe study However it further demonstrates the applicability of theproposed solution to a wide range of disturbances The frequencymeasurements are derived from the raw voltage measurements atCell 2 and Cell 3 nodes using 3-phase voltage transformers A real-time system controller a multi-processor based system with analogto digital sampling and digital pre-filtering at 66666micros is utilizedto process the measurements As in the simulation based studiesthe measurement window to obtain RoCoF remains Tn = 002sThe parameters of the control system remain same as in the pre-vious study The system frequency response subject to generationloss at t = 10s is shown in Fig 7b As can be observed the ADDwas able to identify the imbalance evident from the restoration offrequency much faster than CLFC Therefore demonstration of thefeasibility of ADD operation with real measurements strongly sup-ports the deployment of the proposed control approach in futurepower systems

6 Conclusions

In this paper to improve the speed of response of secondary fre-quency control a novel control has been proposed The controlincorporates a fast acting balance control loop in addition to the sec-ondary control loop The balance control loop incorporates an areadistrubance determiner capable of fast autonomous and definitiveidentification of location of the imbalance The impact of fast anddefinitive identification of location of imbalance is two-fold

bull it ensures unilateral activation of reserves only within the areawhere the imbalance initiated This allows for faster response speedbull it enables use of area power imbalance observed over tie-lines ascontrol input as opposed to conventionally utilized area control errorThis avoids sympathetic activations even at faster response speeds

The proposed enhanced load frequency control further decouples thecontrol effort of fast acting resources from that of conventional gen-erators The control effort to fast acting resources is a combination ofeffort from balance control loop and secondary control loop whilethe control effort to conventional generators is from secondary con-trol loop only This allows for the exploitation of true effectiveness offast acting devices for frequency restoration The analytical assess-ments in the work undertaken provides a high degree of confidencein the aptness of the proposed approach for dynamically changingpower systems while the experimental evaluation demonstrates itsreal-world applicability

Acknowledgment

This work was supported by the European Commission under theFP7 ELECTRA IRP (grant no 609687) Any opinions findings andconclusions or recommendations expressed in this material are those

of the authors and do not necessarily reflect those of the EuropeanCommission

7 References1 lsquoNational Transmission Network Development Planrsquo (Australian Energy

Market Operator (AEMO) 2016) Available from httpwwwaemocomauElectricityNational-Electricity-Market-NEMPlanning-and-forecastingNational-Transmission-Network-Development-Plan

2 Torres LMA Lopes LAC MorAtildean TLA Espinoza CJR lsquoSelf-tuningvirtual synchronous machine A control strategy for energy storage systems to sup-port dynamic frequency controlrsquo IEEE Transactions on Energy Conversion 201429 (4) pp 833ndash840

3 Rakhshani E Remon D Cantarellas AM Rodriguez P lsquoAnalysis of deriva-tive control based virtual inertia in multi-area high-voltage direct current intercon-nected power systemsrsquo IET Generation Transmission Distribution 2016 10 (6)pp 1458ndash1469

4 Alipoor J Miura Y Ise T lsquoPower system stabilization using virtual syn-chronous generator with alternating moment of inertiarsquo IEEE Journal of Emergingand Selected Topics in Power Electronics 2015 3 (2) pp 451ndash458

5 Soni N Doolla S Chandorkar MC lsquoImprovement of transient response inmicrogrids using virtual inertiarsquo IEEE Transactions on Power Delivery 2013 28(3) pp 1830ndash1838

6 lsquoFast Frequency Response in the NEM Working Paperrsquo (Aus-tralian Energy Market Operator (AEMO) 2017) Available fromhttpswwwaemocomau-mediaFilesElectricityNEMSecurityandReliabilityReports2017FFR-Working-Paper---Finalpdf

7 lsquoSystem Operability Framework 2016rsquo (National Grid 2016) Availablefrom httpwww2nationalgridcomUKIndustry-informationFuture-of-EnergySystem-Operability-Framework

8 lsquoRoCoF - An independent analysis on the ability of Generators to ride throughRate of Change of Frequency values up to 2Hzsrsquo (DNV-KEMA for Eir-GridSONI 2013) Available from httpwwweirgridgroupcomsite-fileslibraryEirGridDNV-KEMA_Report_RoCoF_20130208final_pdf

9 lsquoInternational Review of Frequency Control Adaptationrsquo (DGA Consultingreport to Australian Energy Market Operator (AEMO) 2016) Available fromhttpswwwaemocomau-mediaFilesElectricityNEMSecurityandReliabilityReportsFPSS---International-Review-of-FrequencyControlpdf

10 lsquoPolicy P1 Load-Frequency Control and Performancersquo (in Continen-tal Europe Operation Handbook ENTSO-E 2016) Available fromhttpswwwentsoeeufileadminuser_upload_librarypublicationsentsoeOperation_HandbookPolicy_1_finalpdf

11 lsquoBalancing and Frequency Controlrsquo (North American Electric ReliabilityCorporation 2011) Available from httpwwwnerccomdocsocrsNERC20Balancing20and20Frequency20Control20040520111pdf

12 lsquoGuide to Ancillary Services in the National Electricity Mar-ketrsquo (Australian Energy Market Operator (AEMO) 2015)Available from httpswwwaemocomau-mediaFilesPDFGuide-to-Ancillary-Services-in-the-National-Electricity-Marketashx

13 Jin C Lu N Lu S Makarov YV Dougal RA lsquoA coordinating algorithm fordispatching regulation services between slow and fast power regulating resourcesrsquoIEEE Transactions on Smart Grid 2014 5 (2) pp 1043ndash1050

14 Guo Y Xie X Wang B Shi W Dong Y Mou L et al lsquoImproving agcperformance of a coal-fueled generators with mw-level bessrsquo In 2016 IEEE PowerEnergy Society Innovative Smart Grid Technologies Conference (ISGT) ( 2016pp 1ndash5

15 Xie X Guo Y Wang B Dong Y Mou L Xue F lsquoImproving agc perfor-mance of coal-fueled thermal generators using multi-mw scale bess A practicalapplicationrsquo IEEE Transactions on Smart Grid 2018 9 (3) pp 1769ndash1777

16 Zhang F Hu Z Xie X Zhang J Song Y lsquoAssessment of the effectivenessof energy storage resources in the frequency regulation of a single-area powersystemrsquo IEEE Transactions on Power Systems 2017 32 (5) pp 3373ndash3380

17 KintnerMeyer M lsquoRegulatory policy and markets for energy storage in northamericarsquo Proceedings of the IEEE 2014 102 (7) pp 1065ndash1072

18 lsquoFrequency Control by Demand Management (FCDM)rsquo (National Grid 2015)Available from httpwww2nationalgridcomWorkAreaDownloadAssetaspxid=41294

19 Short JA Infield DG Freris LL lsquoStabilization of grid frequency throughdynamic demand controlrsquo IEEE Transactions on Power Systems 2007 22 (3)pp 1284ndash1293

20 Hu J Cao J Guerrero JM Yong T Yu J lsquoImproving frequency stabilitybased on distributed control of multiple load aggregatorsrsquo IEEE Transactions onSmart Grid 2017 8 (4) pp 1553ndash1567

21 Variani MH Tomsovic K lsquoDistributed automatic generation control usingflatness-based approach for high penetration of wind generationrsquo IEEE Transac-tions on Power Systems 2013 28 (3) pp 3002ndash3009

22 Xie P Li Y Zhu L Shi D Duan X lsquoSupplementary automatic genera-tion control using controllable energy storage in electric vehicle battery swappingstationsrsquo IET Generation Transmission Distribution 2016 10 (4) pp 1107ndash1116

23 Keyhani A Chatterjee A lsquoAutomatic generation control structure for smartpower gridsrsquo IEEE Transactions on Smart Grid 2012 3 (3) pp 1310ndash1316

24 Kok K Widergren S lsquoA society of devices Integrating intelligent distributedresources with transactive energyrsquo IEEE Power and Energy Magazine 2016 14(3) pp 34ndash45

25 R LGraham Knuth DE Patashnik O lsquoConcrete Mathematics A Foun-dation for Computer Sciencersquo (Addison-Wesley Publishing Company1990) Available from httpswwwcsientuedutw~r97002tempConcrete20Mathematics202epdf

IET Research Journals pp 1ndash10ccopy The Institution of Engineering and Technology 2015 9

5 Performance Evaluation

In this section the applicability of the proposed ELFC is evaluatedby means of real-time simulations and experiments within a state-of-the-art smart grid laboratory A three step approach for validation isadopted as follows

1 Simulation and analysis The feasibility of the approach is veri-fied by means of real-time simulations The response of the systemsubject to reference disturbance is evaluated followed by a sensitiv-ity analysis with respect to system inertia and size of disturbanceThis is followed by bringing forth the advantages offered by the keyfeatures of the BCL ie the ADD and use of tie-line power flow(∆P tie

i ) as control input instead of the ACE2 Controller Hardware-in-the-Loop validation Such a validationserves two main purposes (i) the validation of the controller underrealistic delays and (ii) the prototyping of the proposed controlwithin an actual hardware controller to demonstrate its real-worldapplication3 Controller and Power Hardware-in-the-Loop validation The per-formance improvement of the proposed ELFC over CLFC has beenachieved by means of incorporation of BCL that relies on ADDfor a faster disturbance detection The disturbance detection utilizesRoCoF that is regarded as a very noisy measurement Any con-trol that relies on RoCoF should therefore be validated by means ofreal measurements enabled by controller and power hardware-in-the-loop implementation Therefore the ADD of the proposed control isvalidated in this way

In all the above cases the performance of the proposed ELFC iscompared to that of CLFC To aid the assessment two key indica-tors are defined namely (i) frequency restoration time (T rest) and(ii) relative frequency overshoot (∆fovr) To asses the T rest of thecontrollers an error margin ε = 001 is defined such that T rest isthe time interval between the initiation of the disturbance to the pointwhen |∆f |= |fmeas minus fnom|lt ε and can be represented as

T rest = T rest isin R forall t gt T rest |∆f |le ε (15)

Defining the minimum and maximum frequency after a disturbanceas fmax and fmin respectively the relative overshoot is calculatedas

∆fovr =

∣∣∣∣fmax minus fnomfnom minus fmin

∣∣∣∣ (16)

As the GB power system is a relatively strongly coupled networkie a stiff network therefore the same frequency characteristics areobserved throughout the network [39] For simplicity and clarityof the figures only the frequency of the area with a disturbance isplotted for the analysis that follows

51 Simulation and analysis

511 Response to reference disturbance The system fre-quency responses subject to a generation loss at t = 10s and a loadloss at t = 110s for CLFC and ELFC are presented in Fig 4a Therestoration time for the CLFC is 83s with no overshoot The pro-posed ELFC by means of fast and accurate detection of disturbancelocation contributes to improving frequency nadir (frequency zenithfor load loss) and leads to a faster restoration time With a restora-tion time of 352s the proposed ELFC is twice as fast but presentsan overshoot of 35 An acceptable frequency response of a sys-tem subject to a disturbance is defined by an exponentially decayingfunction H(s) = fnom plusmnAeminus

1T t referred to as the trumpet curve

set as a requirement by the European Network of Transmission Sys-tem Operators for Electricity [31] Considering the response of theconventional control as reference trumpet curve parameters are cal-culated as A = 031Hz and T = 11s As can be observed from Fig4a the response of the proposed ELFC with a 35 overshoot iswell within the bound set by the trumpet curve The overshoot of theELFC can be abated and the response further improved by means

0 50 100 150 200

Time (s)

497

498

499

50

501

502

503

Fre

qu

en

cy

(H

z)

ELFC

CLFC

Trumpet Curve

Trumpet Curve

(a) Frequency response

0 50 100 150 200-005

0

005

AD

D O

utp

ut

Pti

e0 50 100 150 200

Time (s)

0

002

004

006

To

tal

Re

gu

lati

on

Po

we

r (p

u)

10 101 102 103 104

-004

0

Other LFC

ADD Outputs

LFC 2 ADD

Output

LFC 2

Other

LFCs

(b) ADD output and total regulation power for ELFC

microStep

microInertia

701

80

09 2

Re

sto

rati

on

Tim

e (

s)

90

CLFC

08 15

100

071

06 microStep

microInertia

25

1

30

09 2

Re

sto

rati

on

Tim

e (

s)

35

ELFC

08 15

40

071

0630

40

50

60

70

80

90

100

(c) Sensitivity analysis

Fig 4 Simulation and sensitivity analysis results for performanceevaluation of proposed ELFC

of faster acting PFC where primary response is provided by bat-tery energy storage systems as opposed to CGs as is the case forsimulations undertaken

The output of ADD and the total regulation power ie the sumof the regulation command from the DSAs and the secondary con-trol loop for all the areas subject to the generation loss and loadloss is shown in Fig 4b As can be observed only the output ofLFC 2 ADD is non-zero (and equal to ∆P tie

2 ) consequent to suc-cessful disturbance identification Furthermore this identification iswithin 300ms (sim 250ms) as shown This enables the fast activationof resources through the BCL to be activated unilaterally ie only inarea with disturbance as is evident from the total regulation powershown

512 Sensitivity analysis To further evaluate the performanceof the proposed ELFC a sensitivity analysis for variation in size ofdisturbance and system inertia is undertaken The size of disturbance(P step) is varied by means of a multiplier microstep such that P step =

microstepPstep0 where P step

0 is the reference disturbance magnitudeIn a similar manner a multiplier microinertia scales the system inertiaconstant (H) such that H = microinertiaH0 where H0 = 468s Theresults of the analysis for CLFC and ELFC are presented in Fig 4cAs can be observed the restoration time for CLFC increases with

IET Research Journals pp 1ndash106 ccopy The Institution of Engineering and Technology 2015

(a) CLFC with auxiliary controller-002 0 002 006

-100

-50

0

50

100

Imag

inary

Axis

(seco

nd

s-1

)

(seconds-1)

(seconds

-1)

-003 -001 0 001

-100

-50

0

50

100

Imag

inary

Axis

(seco

nd

s-1

)

(seconds-1)

(seconds

-1)

CLFC ELFC

Real Axis (seconds )-1Real Axis (seconds )-1

TIaux

=50 s

TIaux

=5 s

TIaux

=50 s

TIaux

=5 s TIaux

= 50 s rarr 5 s

TIaux

= 50 s rarr 5 s

(b) Eigenvalue analysis

0 10 20 30 40 50 60 70 80 90 100

Time (s)

497

498

499

50

501

502

Fre

qu

en

cy (

Hz)

Accurate Bias Factor

5-

Error in Bias Factor

5+

Error in Bias Factor

ELFC

Responses

CLFC

Responses

(c) Frequency response

0 10 20 30 40 50 60 70 80 90 100

Time (s)

0

002

004

006

008

01

Reg

ula

tio

n P

ow

er

(pu

)

5-

Error in Bias Factor

5+

Error in Bias Factor

Responses for

ELFC using ACE

Responses for

ELFC using ∆ Ptie

(d) Regulation power

Fig 5 Results for performance evaluation of proposed ELFC

increasing disturbance size but the decreasing system inertia doesnot have a significant impact On the other hand the performanceof proposed ELFC is significantly different the restoration time ofwhich is not affected by the increase in disturbance size while atthe same time a decrease in system inertia leads to faster restora-tion times Therefore the proposed ELFC is more robust to varyingdisturbance sizes and system characteristics a necessity for futurepower systems

With variations in disturbance size and system inertia the relativeovershoot for the proposed ELFC is always under 40 and withinthe bounds set by the respective trumpet curve While the CLFCexhibits no overshooting characteristic with decreasing system iner-tia the frequency nadir significantly decreases The proposed ELFCimproves the frequency nadir for all the cases assessed

513 Importance of ADD The objective of this sub-section isto demonstrate that the improvement in the performance of the pro-posed control is not only due to the incorporation of additional fasterauxiliary controller but the fact that the identification of the locationof the disturbance does play a very important role Therefore theimportance of ADD is demonstrated by means of conducting small-signal stability analysis on the test power system The two cases forcomparison are (i) the proposed ELFC control and (ii) the CLFCincorporating the auxiliary controller only as shown in Fig 5a Thestate-space model of M-area interconnected power system with BCL(as shown in Fig 2) can be represented as [40]

x = Ax+Bu+ Fw

y = Cx(17)

with x = [x1 x2 x3 x4 x5]T

u = [u1 u2]T

w = [∆PL1 ∆PLM

]T(18)

where x is the state vector u is the control vector andw is the disturbance vector The internal states can be rep-resented as x1 = [∆f1 ∆fM ] x2 = [∆Pm1 ∆PmM ]x3 =

[∆P tie

1 ∆P tieM

] x4 =

[intACE1

intACEM

]and

x5 =[intd1

intdM] The internal inputs can be represented

as u1 = [∆Pc1 ∆PcM ] and u2 =[∆PDSA

1 ∆PDSAM

]

The coefficient matrices A B and C can be found in [40] Thesystem is linearized around fnom in steady-state The input to thesystem is the reference disturbance in area 2 and the output is thesystem frequency

The Eigenvalues for the two cases are obtained by varying theTIaux from 50s to 20s in steps of 10s and then to 5s in steps of 5sIn Fig 5b the pole-zero map for the CLFC (left) and the proposedELFC (right) have been presented Only the states that are impactedby TIaux are shown for clearer representation As can be observedfrom Fig 5b for CLFC as the value of TIaux is reduced the polesmove towards the imaginary axis For TIaux = 10s the poles crossthe imaginary axis representing an unstable system However forthe proposed ELFC where the ADD ensures unilateral reserve acti-vations only within the area where the disturbance has occurred thepoles are very slightly impacted by the reduction in TIaux whichtherefore does not deteriorate the stability of the system

514 Choosing ∆P tie over ACE This sub-section presents ajustification for the use of ∆P tie within BCL as opposed to the con-ventionally used ACE Consider the frequency responses presentedin Fig 5c with frequency bias factor miscalculation of 5 for boththe CLFC and proposed ELFC As can be observed a 5 miscalcu-lation of β results in a plusmn17s variation in T set for the CLFC whilehas no significant impact within the proposed ELFC β is a combina-tion of the droop set for the area and the damping offered by the loadwithin the area and it is not uncommon for errors in its calculationLarger generating units within the network are bound by the gridcode to provide a set percentage response to a variation in frequencyHowever if the units do not abide by the set droop this will lead tomiscalculation of β Furthermore presently it is normal practice todetermine β on a yearly basis [11] In future systems it is expectedthat the characteristics of the network (eg system inertia) will varyvastly even within a single day It is also expected that the differencebetween the peak load and the base load will increase significantly[7] thereby resulting in the load offering a different damping effectat different times Therefore the proposed ELFC utilizing ∆P tie ismore resilient to changes expected within the future power system

IET Research Journals pp 1ndash10ccopy The Institution of Engineering and Technology 2015 7

1 15 2 25 3 35 4 45

Delay (s)

0

1000

2000

3000

4000

5000

Sam

ple

s

Uniform Distribution

Gaussian Distribution

(a) Delay distributions under consideration

(b) Controller hardware-in-the-loop setup

0 10 20 30 40 50 60 70 80 90 100

Time (s)

4975

498

4985

499

4995

50

5005

501

Fre

qu

en

cy (

Hz)

No delay

Gaussian Distribution Delay

Uniform Distribution Delay

(c) Frequency response with and without delays

Fig 6 Setup and results for controller hardware-in-the-loop imple-mentation

Furthermore if ACE was instead utilized within the BCL of theproposed scheme although the activations of reserves are guaran-teed to be unilateral within the area where the disturbance has beendetected the miscalculation of bias factor would lead to nuisanceactivations as shown in Fig 5d It should be noted that utilizing∆P tie avoids any nuisance activations

52 Controller Hardware-in-the-Loop Validation

A detailed study on experimental validation of ancillary serviceprovision capability of DSAs controlling a portfolio of highly dis-tributed demand side devices has been undertaken in [41] Thestudy identifies that validation of any control relying on DSAs forprovision of time critical ancillary services should incorporate real-istic communication delays to demonstrate robustness The resultspresented in previous sub-sections assume a fixed communicationsdelay however the communication delay referred to in this paperis the time it takes for a control command from the aggregator toreach the participating device and is highly dependent upon the com-munications architecture employed by the potential demand sideaggregator

The performance of the communications networks that connect acontrol centre to the end devices for demand side applications hasbeen analyzed in [42] and the results show that the latency expected

Table 1 Test Metrics

Delay TRest (s) ∆fovr ()

No delay 2416 287Gaussian distribution delay 4237 44Uniform distribution delay 423 436

(a) controller and power hardware-in-the-loop setup

0 10 20 30 40 50 60 70 80 90 100

Time (s)

498

499

50

501

Fre

qu

en

cy (

Hz)

CLFC

ELFC

(b) Frequency response

Fig 7 Setup and results for controller and power hardware-in-the-loop implementation

is between 1minus 45s The analysis presents an estimation of delay butdoes not provide a mean value for the delay nor a standard deviationThe variation in delay is mostly associated to the number of hopsa message packet makes in the neighborhood area network beforeit reaches the device As a constant delay is highly unlikely in areal-world scenario for this work two types of distributions are con-sidered as shown in Fig 6a a uniform distribution and a Gaussiandistribution where the mean value micro has been chosen as 275s (aver-age of minimum and maximum latency) with a standard deviation ofσ of 03s

A controller hardware-in-the-loop implementation as presentedin Fig 6b is utilized The five-areas of the GB power system aresimulated within the digital real-time simulator the proposed ELFCis implemented within a Beckhoff real-time controller while a sim-ple transactive energy based DSA runs on the host PC The proposedELFC is run at a time step of 10ms with TIaux = 5s while all othercontrol parameters remain the same as in the simulation study Theresponse of the system subject to the reference imbalance is pre-sented in Fig 6c and test metrics in Table 1 As can be observedthe response of the proposed ELFC incorporating delays charac-terized for demand side applications is stable and satisfactory andthere is not much difference in response with two different delaydistributions The deployment of the control on dedicated controllerhardware enabled by rapid prototyping further demonstrates theviability of the control for real-world implementations

IET Research Journals pp 1ndash108 ccopy The Institution of Engineering and Technology 2015

53 Controller and Power Hardware-in-the-Loop

The controller and power hardware-in-the-loop setup is shown inFig 7a As can be observed two areas of the GB network arerepresented by Cell 2 and Cell 3 of the Dynamic Power SystemsLaboratory [43] while the remaining three areas are simulated withinthe digital real-time simulator To ensure stability of the setup andto be able to represent a large portion of the GB network thecurrents are scaled by a factor of k = 08times 106 A detailed descrip-tion and process of initialization and synchronization of the powerhardware-in-the-loop setup is found in [44]

The system is subject to disturbance in LFC area 2 (Cell 2 inhardware Fig 7a) a loss of generation with magnitude of 800MW(1000W in hardware) A smaller magnitude of disturbance in exper-imental evaluation compared to the reference imbalance magnitudeis due to the stability of the setup which put a limit on the sparepower capacity reserved for imbalance event As the objective ofpower hardware-in-the-loop evaluation is to demonstrate the capa-bility of the proposed control to operate under real measurementsthe magnitude of the disturbance does not impact the integrity ofthe study However it further demonstrates the applicability of theproposed solution to a wide range of disturbances The frequencymeasurements are derived from the raw voltage measurements atCell 2 and Cell 3 nodes using 3-phase voltage transformers A real-time system controller a multi-processor based system with analogto digital sampling and digital pre-filtering at 66666micros is utilizedto process the measurements As in the simulation based studiesthe measurement window to obtain RoCoF remains Tn = 002sThe parameters of the control system remain same as in the pre-vious study The system frequency response subject to generationloss at t = 10s is shown in Fig 7b As can be observed the ADDwas able to identify the imbalance evident from the restoration offrequency much faster than CLFC Therefore demonstration of thefeasibility of ADD operation with real measurements strongly sup-ports the deployment of the proposed control approach in futurepower systems

6 Conclusions

In this paper to improve the speed of response of secondary fre-quency control a novel control has been proposed The controlincorporates a fast acting balance control loop in addition to the sec-ondary control loop The balance control loop incorporates an areadistrubance determiner capable of fast autonomous and definitiveidentification of location of the imbalance The impact of fast anddefinitive identification of location of imbalance is two-fold

bull it ensures unilateral activation of reserves only within the areawhere the imbalance initiated This allows for faster response speedbull it enables use of area power imbalance observed over tie-lines ascontrol input as opposed to conventionally utilized area control errorThis avoids sympathetic activations even at faster response speeds

The proposed enhanced load frequency control further decouples thecontrol effort of fast acting resources from that of conventional gen-erators The control effort to fast acting resources is a combination ofeffort from balance control loop and secondary control loop whilethe control effort to conventional generators is from secondary con-trol loop only This allows for the exploitation of true effectiveness offast acting devices for frequency restoration The analytical assess-ments in the work undertaken provides a high degree of confidencein the aptness of the proposed approach for dynamically changingpower systems while the experimental evaluation demonstrates itsreal-world applicability

Acknowledgment

This work was supported by the European Commission under theFP7 ELECTRA IRP (grant no 609687) Any opinions findings andconclusions or recommendations expressed in this material are those

of the authors and do not necessarily reflect those of the EuropeanCommission

7 References1 lsquoNational Transmission Network Development Planrsquo (Australian Energy

Market Operator (AEMO) 2016) Available from httpwwwaemocomauElectricityNational-Electricity-Market-NEMPlanning-and-forecastingNational-Transmission-Network-Development-Plan

2 Torres LMA Lopes LAC MorAtildean TLA Espinoza CJR lsquoSelf-tuningvirtual synchronous machine A control strategy for energy storage systems to sup-port dynamic frequency controlrsquo IEEE Transactions on Energy Conversion 201429 (4) pp 833ndash840

3 Rakhshani E Remon D Cantarellas AM Rodriguez P lsquoAnalysis of deriva-tive control based virtual inertia in multi-area high-voltage direct current intercon-nected power systemsrsquo IET Generation Transmission Distribution 2016 10 (6)pp 1458ndash1469

4 Alipoor J Miura Y Ise T lsquoPower system stabilization using virtual syn-chronous generator with alternating moment of inertiarsquo IEEE Journal of Emergingand Selected Topics in Power Electronics 2015 3 (2) pp 451ndash458

5 Soni N Doolla S Chandorkar MC lsquoImprovement of transient response inmicrogrids using virtual inertiarsquo IEEE Transactions on Power Delivery 2013 28(3) pp 1830ndash1838

6 lsquoFast Frequency Response in the NEM Working Paperrsquo (Aus-tralian Energy Market Operator (AEMO) 2017) Available fromhttpswwwaemocomau-mediaFilesElectricityNEMSecurityandReliabilityReports2017FFR-Working-Paper---Finalpdf

7 lsquoSystem Operability Framework 2016rsquo (National Grid 2016) Availablefrom httpwww2nationalgridcomUKIndustry-informationFuture-of-EnergySystem-Operability-Framework

8 lsquoRoCoF - An independent analysis on the ability of Generators to ride throughRate of Change of Frequency values up to 2Hzsrsquo (DNV-KEMA for Eir-GridSONI 2013) Available from httpwwweirgridgroupcomsite-fileslibraryEirGridDNV-KEMA_Report_RoCoF_20130208final_pdf

9 lsquoInternational Review of Frequency Control Adaptationrsquo (DGA Consultingreport to Australian Energy Market Operator (AEMO) 2016) Available fromhttpswwwaemocomau-mediaFilesElectricityNEMSecurityandReliabilityReportsFPSS---International-Review-of-FrequencyControlpdf

10 lsquoPolicy P1 Load-Frequency Control and Performancersquo (in Continen-tal Europe Operation Handbook ENTSO-E 2016) Available fromhttpswwwentsoeeufileadminuser_upload_librarypublicationsentsoeOperation_HandbookPolicy_1_finalpdf

11 lsquoBalancing and Frequency Controlrsquo (North American Electric ReliabilityCorporation 2011) Available from httpwwwnerccomdocsocrsNERC20Balancing20and20Frequency20Control20040520111pdf

12 lsquoGuide to Ancillary Services in the National Electricity Mar-ketrsquo (Australian Energy Market Operator (AEMO) 2015)Available from httpswwwaemocomau-mediaFilesPDFGuide-to-Ancillary-Services-in-the-National-Electricity-Marketashx

13 Jin C Lu N Lu S Makarov YV Dougal RA lsquoA coordinating algorithm fordispatching regulation services between slow and fast power regulating resourcesrsquoIEEE Transactions on Smart Grid 2014 5 (2) pp 1043ndash1050

14 Guo Y Xie X Wang B Shi W Dong Y Mou L et al lsquoImproving agcperformance of a coal-fueled generators with mw-level bessrsquo In 2016 IEEE PowerEnergy Society Innovative Smart Grid Technologies Conference (ISGT) ( 2016pp 1ndash5

15 Xie X Guo Y Wang B Dong Y Mou L Xue F lsquoImproving agc perfor-mance of coal-fueled thermal generators using multi-mw scale bess A practicalapplicationrsquo IEEE Transactions on Smart Grid 2018 9 (3) pp 1769ndash1777

16 Zhang F Hu Z Xie X Zhang J Song Y lsquoAssessment of the effectivenessof energy storage resources in the frequency regulation of a single-area powersystemrsquo IEEE Transactions on Power Systems 2017 32 (5) pp 3373ndash3380

17 KintnerMeyer M lsquoRegulatory policy and markets for energy storage in northamericarsquo Proceedings of the IEEE 2014 102 (7) pp 1065ndash1072

18 lsquoFrequency Control by Demand Management (FCDM)rsquo (National Grid 2015)Available from httpwww2nationalgridcomWorkAreaDownloadAssetaspxid=41294

19 Short JA Infield DG Freris LL lsquoStabilization of grid frequency throughdynamic demand controlrsquo IEEE Transactions on Power Systems 2007 22 (3)pp 1284ndash1293

20 Hu J Cao J Guerrero JM Yong T Yu J lsquoImproving frequency stabilitybased on distributed control of multiple load aggregatorsrsquo IEEE Transactions onSmart Grid 2017 8 (4) pp 1553ndash1567

21 Variani MH Tomsovic K lsquoDistributed automatic generation control usingflatness-based approach for high penetration of wind generationrsquo IEEE Transac-tions on Power Systems 2013 28 (3) pp 3002ndash3009

22 Xie P Li Y Zhu L Shi D Duan X lsquoSupplementary automatic genera-tion control using controllable energy storage in electric vehicle battery swappingstationsrsquo IET Generation Transmission Distribution 2016 10 (4) pp 1107ndash1116

23 Keyhani A Chatterjee A lsquoAutomatic generation control structure for smartpower gridsrsquo IEEE Transactions on Smart Grid 2012 3 (3) pp 1310ndash1316

24 Kok K Widergren S lsquoA society of devices Integrating intelligent distributedresources with transactive energyrsquo IEEE Power and Energy Magazine 2016 14(3) pp 34ndash45

25 R LGraham Knuth DE Patashnik O lsquoConcrete Mathematics A Foun-dation for Computer Sciencersquo (Addison-Wesley Publishing Company1990) Available from httpswwwcsientuedutw~r97002tempConcrete20Mathematics202epdf

IET Research Journals pp 1ndash10ccopy The Institution of Engineering and Technology 2015 9

(a) CLFC with auxiliary controller-002 0 002 006

-100

-50

0

50

100

Imag

inary

Axis

(seco

nd

s-1

)

(seconds-1)

(seconds

-1)

-003 -001 0 001

-100

-50

0

50

100

Imag

inary

Axis

(seco

nd

s-1

)

(seconds-1)

(seconds

-1)

CLFC ELFC

Real Axis (seconds )-1Real Axis (seconds )-1

TIaux

=50 s

TIaux

=5 s

TIaux

=50 s

TIaux

=5 s TIaux

= 50 s rarr 5 s

TIaux

= 50 s rarr 5 s

(b) Eigenvalue analysis

0 10 20 30 40 50 60 70 80 90 100

Time (s)

497

498

499

50

501

502

Fre

qu

en

cy (

Hz)

Accurate Bias Factor

5-

Error in Bias Factor

5+

Error in Bias Factor

ELFC

Responses

CLFC

Responses

(c) Frequency response

0 10 20 30 40 50 60 70 80 90 100

Time (s)

0

002

004

006

008

01

Reg

ula

tio

n P

ow

er

(pu

)

5-

Error in Bias Factor

5+

Error in Bias Factor

Responses for

ELFC using ACE

Responses for

ELFC using ∆ Ptie

(d) Regulation power

Fig 5 Results for performance evaluation of proposed ELFC

increasing disturbance size but the decreasing system inertia doesnot have a significant impact On the other hand the performanceof proposed ELFC is significantly different the restoration time ofwhich is not affected by the increase in disturbance size while atthe same time a decrease in system inertia leads to faster restora-tion times Therefore the proposed ELFC is more robust to varyingdisturbance sizes and system characteristics a necessity for futurepower systems

With variations in disturbance size and system inertia the relativeovershoot for the proposed ELFC is always under 40 and withinthe bounds set by the respective trumpet curve While the CLFCexhibits no overshooting characteristic with decreasing system iner-tia the frequency nadir significantly decreases The proposed ELFCimproves the frequency nadir for all the cases assessed

513 Importance of ADD The objective of this sub-section isto demonstrate that the improvement in the performance of the pro-posed control is not only due to the incorporation of additional fasterauxiliary controller but the fact that the identification of the locationof the disturbance does play a very important role Therefore theimportance of ADD is demonstrated by means of conducting small-signal stability analysis on the test power system The two cases forcomparison are (i) the proposed ELFC control and (ii) the CLFCincorporating the auxiliary controller only as shown in Fig 5a Thestate-space model of M-area interconnected power system with BCL(as shown in Fig 2) can be represented as [40]

x = Ax+Bu+ Fw

y = Cx(17)

with x = [x1 x2 x3 x4 x5]T

u = [u1 u2]T

w = [∆PL1 ∆PLM

]T(18)

where x is the state vector u is the control vector andw is the disturbance vector The internal states can be rep-resented as x1 = [∆f1 ∆fM ] x2 = [∆Pm1 ∆PmM ]x3 =

[∆P tie

1 ∆P tieM

] x4 =

[intACE1

intACEM

]and

x5 =[intd1

intdM] The internal inputs can be represented

as u1 = [∆Pc1 ∆PcM ] and u2 =[∆PDSA

1 ∆PDSAM

]

The coefficient matrices A B and C can be found in [40] Thesystem is linearized around fnom in steady-state The input to thesystem is the reference disturbance in area 2 and the output is thesystem frequency

The Eigenvalues for the two cases are obtained by varying theTIaux from 50s to 20s in steps of 10s and then to 5s in steps of 5sIn Fig 5b the pole-zero map for the CLFC (left) and the proposedELFC (right) have been presented Only the states that are impactedby TIaux are shown for clearer representation As can be observedfrom Fig 5b for CLFC as the value of TIaux is reduced the polesmove towards the imaginary axis For TIaux = 10s the poles crossthe imaginary axis representing an unstable system However forthe proposed ELFC where the ADD ensures unilateral reserve acti-vations only within the area where the disturbance has occurred thepoles are very slightly impacted by the reduction in TIaux whichtherefore does not deteriorate the stability of the system

514 Choosing ∆P tie over ACE This sub-section presents ajustification for the use of ∆P tie within BCL as opposed to the con-ventionally used ACE Consider the frequency responses presentedin Fig 5c with frequency bias factor miscalculation of 5 for boththe CLFC and proposed ELFC As can be observed a 5 miscalcu-lation of β results in a plusmn17s variation in T set for the CLFC whilehas no significant impact within the proposed ELFC β is a combina-tion of the droop set for the area and the damping offered by the loadwithin the area and it is not uncommon for errors in its calculationLarger generating units within the network are bound by the gridcode to provide a set percentage response to a variation in frequencyHowever if the units do not abide by the set droop this will lead tomiscalculation of β Furthermore presently it is normal practice todetermine β on a yearly basis [11] In future systems it is expectedthat the characteristics of the network (eg system inertia) will varyvastly even within a single day It is also expected that the differencebetween the peak load and the base load will increase significantly[7] thereby resulting in the load offering a different damping effectat different times Therefore the proposed ELFC utilizing ∆P tie ismore resilient to changes expected within the future power system

IET Research Journals pp 1ndash10ccopy The Institution of Engineering and Technology 2015 7

1 15 2 25 3 35 4 45

Delay (s)

0

1000

2000

3000

4000

5000

Sam

ple

s

Uniform Distribution

Gaussian Distribution

(a) Delay distributions under consideration

(b) Controller hardware-in-the-loop setup

0 10 20 30 40 50 60 70 80 90 100

Time (s)

4975

498

4985

499

4995

50

5005

501

Fre

qu

en

cy (

Hz)

No delay

Gaussian Distribution Delay

Uniform Distribution Delay

(c) Frequency response with and without delays

Fig 6 Setup and results for controller hardware-in-the-loop imple-mentation

Furthermore if ACE was instead utilized within the BCL of theproposed scheme although the activations of reserves are guaran-teed to be unilateral within the area where the disturbance has beendetected the miscalculation of bias factor would lead to nuisanceactivations as shown in Fig 5d It should be noted that utilizing∆P tie avoids any nuisance activations

52 Controller Hardware-in-the-Loop Validation

A detailed study on experimental validation of ancillary serviceprovision capability of DSAs controlling a portfolio of highly dis-tributed demand side devices has been undertaken in [41] Thestudy identifies that validation of any control relying on DSAs forprovision of time critical ancillary services should incorporate real-istic communication delays to demonstrate robustness The resultspresented in previous sub-sections assume a fixed communicationsdelay however the communication delay referred to in this paperis the time it takes for a control command from the aggregator toreach the participating device and is highly dependent upon the com-munications architecture employed by the potential demand sideaggregator

The performance of the communications networks that connect acontrol centre to the end devices for demand side applications hasbeen analyzed in [42] and the results show that the latency expected

Table 1 Test Metrics

Delay TRest (s) ∆fovr ()

No delay 2416 287Gaussian distribution delay 4237 44Uniform distribution delay 423 436

(a) controller and power hardware-in-the-loop setup

0 10 20 30 40 50 60 70 80 90 100

Time (s)

498

499

50

501

Fre

qu

en

cy (

Hz)

CLFC

ELFC

(b) Frequency response

Fig 7 Setup and results for controller and power hardware-in-the-loop implementation

is between 1minus 45s The analysis presents an estimation of delay butdoes not provide a mean value for the delay nor a standard deviationThe variation in delay is mostly associated to the number of hopsa message packet makes in the neighborhood area network beforeit reaches the device As a constant delay is highly unlikely in areal-world scenario for this work two types of distributions are con-sidered as shown in Fig 6a a uniform distribution and a Gaussiandistribution where the mean value micro has been chosen as 275s (aver-age of minimum and maximum latency) with a standard deviation ofσ of 03s

A controller hardware-in-the-loop implementation as presentedin Fig 6b is utilized The five-areas of the GB power system aresimulated within the digital real-time simulator the proposed ELFCis implemented within a Beckhoff real-time controller while a sim-ple transactive energy based DSA runs on the host PC The proposedELFC is run at a time step of 10ms with TIaux = 5s while all othercontrol parameters remain the same as in the simulation study Theresponse of the system subject to the reference imbalance is pre-sented in Fig 6c and test metrics in Table 1 As can be observedthe response of the proposed ELFC incorporating delays charac-terized for demand side applications is stable and satisfactory andthere is not much difference in response with two different delaydistributions The deployment of the control on dedicated controllerhardware enabled by rapid prototyping further demonstrates theviability of the control for real-world implementations

IET Research Journals pp 1ndash108 ccopy The Institution of Engineering and Technology 2015

53 Controller and Power Hardware-in-the-Loop

The controller and power hardware-in-the-loop setup is shown inFig 7a As can be observed two areas of the GB network arerepresented by Cell 2 and Cell 3 of the Dynamic Power SystemsLaboratory [43] while the remaining three areas are simulated withinthe digital real-time simulator To ensure stability of the setup andto be able to represent a large portion of the GB network thecurrents are scaled by a factor of k = 08times 106 A detailed descrip-tion and process of initialization and synchronization of the powerhardware-in-the-loop setup is found in [44]

The system is subject to disturbance in LFC area 2 (Cell 2 inhardware Fig 7a) a loss of generation with magnitude of 800MW(1000W in hardware) A smaller magnitude of disturbance in exper-imental evaluation compared to the reference imbalance magnitudeis due to the stability of the setup which put a limit on the sparepower capacity reserved for imbalance event As the objective ofpower hardware-in-the-loop evaluation is to demonstrate the capa-bility of the proposed control to operate under real measurementsthe magnitude of the disturbance does not impact the integrity ofthe study However it further demonstrates the applicability of theproposed solution to a wide range of disturbances The frequencymeasurements are derived from the raw voltage measurements atCell 2 and Cell 3 nodes using 3-phase voltage transformers A real-time system controller a multi-processor based system with analogto digital sampling and digital pre-filtering at 66666micros is utilizedto process the measurements As in the simulation based studiesthe measurement window to obtain RoCoF remains Tn = 002sThe parameters of the control system remain same as in the pre-vious study The system frequency response subject to generationloss at t = 10s is shown in Fig 7b As can be observed the ADDwas able to identify the imbalance evident from the restoration offrequency much faster than CLFC Therefore demonstration of thefeasibility of ADD operation with real measurements strongly sup-ports the deployment of the proposed control approach in futurepower systems

6 Conclusions

In this paper to improve the speed of response of secondary fre-quency control a novel control has been proposed The controlincorporates a fast acting balance control loop in addition to the sec-ondary control loop The balance control loop incorporates an areadistrubance determiner capable of fast autonomous and definitiveidentification of location of the imbalance The impact of fast anddefinitive identification of location of imbalance is two-fold

bull it ensures unilateral activation of reserves only within the areawhere the imbalance initiated This allows for faster response speedbull it enables use of area power imbalance observed over tie-lines ascontrol input as opposed to conventionally utilized area control errorThis avoids sympathetic activations even at faster response speeds

The proposed enhanced load frequency control further decouples thecontrol effort of fast acting resources from that of conventional gen-erators The control effort to fast acting resources is a combination ofeffort from balance control loop and secondary control loop whilethe control effort to conventional generators is from secondary con-trol loop only This allows for the exploitation of true effectiveness offast acting devices for frequency restoration The analytical assess-ments in the work undertaken provides a high degree of confidencein the aptness of the proposed approach for dynamically changingpower systems while the experimental evaluation demonstrates itsreal-world applicability

Acknowledgment

This work was supported by the European Commission under theFP7 ELECTRA IRP (grant no 609687) Any opinions findings andconclusions or recommendations expressed in this material are those

of the authors and do not necessarily reflect those of the EuropeanCommission

7 References1 lsquoNational Transmission Network Development Planrsquo (Australian Energy

Market Operator (AEMO) 2016) Available from httpwwwaemocomauElectricityNational-Electricity-Market-NEMPlanning-and-forecastingNational-Transmission-Network-Development-Plan

2 Torres LMA Lopes LAC MorAtildean TLA Espinoza CJR lsquoSelf-tuningvirtual synchronous machine A control strategy for energy storage systems to sup-port dynamic frequency controlrsquo IEEE Transactions on Energy Conversion 201429 (4) pp 833ndash840

3 Rakhshani E Remon D Cantarellas AM Rodriguez P lsquoAnalysis of deriva-tive control based virtual inertia in multi-area high-voltage direct current intercon-nected power systemsrsquo IET Generation Transmission Distribution 2016 10 (6)pp 1458ndash1469

4 Alipoor J Miura Y Ise T lsquoPower system stabilization using virtual syn-chronous generator with alternating moment of inertiarsquo IEEE Journal of Emergingand Selected Topics in Power Electronics 2015 3 (2) pp 451ndash458

5 Soni N Doolla S Chandorkar MC lsquoImprovement of transient response inmicrogrids using virtual inertiarsquo IEEE Transactions on Power Delivery 2013 28(3) pp 1830ndash1838

6 lsquoFast Frequency Response in the NEM Working Paperrsquo (Aus-tralian Energy Market Operator (AEMO) 2017) Available fromhttpswwwaemocomau-mediaFilesElectricityNEMSecurityandReliabilityReports2017FFR-Working-Paper---Finalpdf

7 lsquoSystem Operability Framework 2016rsquo (National Grid 2016) Availablefrom httpwww2nationalgridcomUKIndustry-informationFuture-of-EnergySystem-Operability-Framework

8 lsquoRoCoF - An independent analysis on the ability of Generators to ride throughRate of Change of Frequency values up to 2Hzsrsquo (DNV-KEMA for Eir-GridSONI 2013) Available from httpwwweirgridgroupcomsite-fileslibraryEirGridDNV-KEMA_Report_RoCoF_20130208final_pdf

9 lsquoInternational Review of Frequency Control Adaptationrsquo (DGA Consultingreport to Australian Energy Market Operator (AEMO) 2016) Available fromhttpswwwaemocomau-mediaFilesElectricityNEMSecurityandReliabilityReportsFPSS---International-Review-of-FrequencyControlpdf

10 lsquoPolicy P1 Load-Frequency Control and Performancersquo (in Continen-tal Europe Operation Handbook ENTSO-E 2016) Available fromhttpswwwentsoeeufileadminuser_upload_librarypublicationsentsoeOperation_HandbookPolicy_1_finalpdf

11 lsquoBalancing and Frequency Controlrsquo (North American Electric ReliabilityCorporation 2011) Available from httpwwwnerccomdocsocrsNERC20Balancing20and20Frequency20Control20040520111pdf

12 lsquoGuide to Ancillary Services in the National Electricity Mar-ketrsquo (Australian Energy Market Operator (AEMO) 2015)Available from httpswwwaemocomau-mediaFilesPDFGuide-to-Ancillary-Services-in-the-National-Electricity-Marketashx

13 Jin C Lu N Lu S Makarov YV Dougal RA lsquoA coordinating algorithm fordispatching regulation services between slow and fast power regulating resourcesrsquoIEEE Transactions on Smart Grid 2014 5 (2) pp 1043ndash1050

14 Guo Y Xie X Wang B Shi W Dong Y Mou L et al lsquoImproving agcperformance of a coal-fueled generators with mw-level bessrsquo In 2016 IEEE PowerEnergy Society Innovative Smart Grid Technologies Conference (ISGT) ( 2016pp 1ndash5

15 Xie X Guo Y Wang B Dong Y Mou L Xue F lsquoImproving agc perfor-mance of coal-fueled thermal generators using multi-mw scale bess A practicalapplicationrsquo IEEE Transactions on Smart Grid 2018 9 (3) pp 1769ndash1777

16 Zhang F Hu Z Xie X Zhang J Song Y lsquoAssessment of the effectivenessof energy storage resources in the frequency regulation of a single-area powersystemrsquo IEEE Transactions on Power Systems 2017 32 (5) pp 3373ndash3380

17 KintnerMeyer M lsquoRegulatory policy and markets for energy storage in northamericarsquo Proceedings of the IEEE 2014 102 (7) pp 1065ndash1072

18 lsquoFrequency Control by Demand Management (FCDM)rsquo (National Grid 2015)Available from httpwww2nationalgridcomWorkAreaDownloadAssetaspxid=41294

19 Short JA Infield DG Freris LL lsquoStabilization of grid frequency throughdynamic demand controlrsquo IEEE Transactions on Power Systems 2007 22 (3)pp 1284ndash1293

20 Hu J Cao J Guerrero JM Yong T Yu J lsquoImproving frequency stabilitybased on distributed control of multiple load aggregatorsrsquo IEEE Transactions onSmart Grid 2017 8 (4) pp 1553ndash1567

21 Variani MH Tomsovic K lsquoDistributed automatic generation control usingflatness-based approach for high penetration of wind generationrsquo IEEE Transac-tions on Power Systems 2013 28 (3) pp 3002ndash3009

22 Xie P Li Y Zhu L Shi D Duan X lsquoSupplementary automatic genera-tion control using controllable energy storage in electric vehicle battery swappingstationsrsquo IET Generation Transmission Distribution 2016 10 (4) pp 1107ndash1116

23 Keyhani A Chatterjee A lsquoAutomatic generation control structure for smartpower gridsrsquo IEEE Transactions on Smart Grid 2012 3 (3) pp 1310ndash1316

24 Kok K Widergren S lsquoA society of devices Integrating intelligent distributedresources with transactive energyrsquo IEEE Power and Energy Magazine 2016 14(3) pp 34ndash45

25 R LGraham Knuth DE Patashnik O lsquoConcrete Mathematics A Foun-dation for Computer Sciencersquo (Addison-Wesley Publishing Company1990) Available from httpswwwcsientuedutw~r97002tempConcrete20Mathematics202epdf

IET Research Journals pp 1ndash10ccopy The Institution of Engineering and Technology 2015 9

1 15 2 25 3 35 4 45

Delay (s)

0

1000

2000

3000

4000

5000

Sam

ple

s

Uniform Distribution

Gaussian Distribution

(a) Delay distributions under consideration

(b) Controller hardware-in-the-loop setup

0 10 20 30 40 50 60 70 80 90 100

Time (s)

4975

498

4985

499

4995

50

5005

501

Fre

qu

en

cy (

Hz)

No delay

Gaussian Distribution Delay

Uniform Distribution Delay

(c) Frequency response with and without delays

Fig 6 Setup and results for controller hardware-in-the-loop imple-mentation

Furthermore if ACE was instead utilized within the BCL of theproposed scheme although the activations of reserves are guaran-teed to be unilateral within the area where the disturbance has beendetected the miscalculation of bias factor would lead to nuisanceactivations as shown in Fig 5d It should be noted that utilizing∆P tie avoids any nuisance activations

52 Controller Hardware-in-the-Loop Validation

A detailed study on experimental validation of ancillary serviceprovision capability of DSAs controlling a portfolio of highly dis-tributed demand side devices has been undertaken in [41] Thestudy identifies that validation of any control relying on DSAs forprovision of time critical ancillary services should incorporate real-istic communication delays to demonstrate robustness The resultspresented in previous sub-sections assume a fixed communicationsdelay however the communication delay referred to in this paperis the time it takes for a control command from the aggregator toreach the participating device and is highly dependent upon the com-munications architecture employed by the potential demand sideaggregator

The performance of the communications networks that connect acontrol centre to the end devices for demand side applications hasbeen analyzed in [42] and the results show that the latency expected

Table 1 Test Metrics

Delay TRest (s) ∆fovr ()

No delay 2416 287Gaussian distribution delay 4237 44Uniform distribution delay 423 436

(a) controller and power hardware-in-the-loop setup

0 10 20 30 40 50 60 70 80 90 100

Time (s)

498

499

50

501

Fre

qu

en

cy (

Hz)

CLFC

ELFC

(b) Frequency response

Fig 7 Setup and results for controller and power hardware-in-the-loop implementation

is between 1minus 45s The analysis presents an estimation of delay butdoes not provide a mean value for the delay nor a standard deviationThe variation in delay is mostly associated to the number of hopsa message packet makes in the neighborhood area network beforeit reaches the device As a constant delay is highly unlikely in areal-world scenario for this work two types of distributions are con-sidered as shown in Fig 6a a uniform distribution and a Gaussiandistribution where the mean value micro has been chosen as 275s (aver-age of minimum and maximum latency) with a standard deviation ofσ of 03s

A controller hardware-in-the-loop implementation as presentedin Fig 6b is utilized The five-areas of the GB power system aresimulated within the digital real-time simulator the proposed ELFCis implemented within a Beckhoff real-time controller while a sim-ple transactive energy based DSA runs on the host PC The proposedELFC is run at a time step of 10ms with TIaux = 5s while all othercontrol parameters remain the same as in the simulation study Theresponse of the system subject to the reference imbalance is pre-sented in Fig 6c and test metrics in Table 1 As can be observedthe response of the proposed ELFC incorporating delays charac-terized for demand side applications is stable and satisfactory andthere is not much difference in response with two different delaydistributions The deployment of the control on dedicated controllerhardware enabled by rapid prototyping further demonstrates theviability of the control for real-world implementations

IET Research Journals pp 1ndash108 ccopy The Institution of Engineering and Technology 2015

53 Controller and Power Hardware-in-the-Loop

The controller and power hardware-in-the-loop setup is shown inFig 7a As can be observed two areas of the GB network arerepresented by Cell 2 and Cell 3 of the Dynamic Power SystemsLaboratory [43] while the remaining three areas are simulated withinthe digital real-time simulator To ensure stability of the setup andto be able to represent a large portion of the GB network thecurrents are scaled by a factor of k = 08times 106 A detailed descrip-tion and process of initialization and synchronization of the powerhardware-in-the-loop setup is found in [44]

The system is subject to disturbance in LFC area 2 (Cell 2 inhardware Fig 7a) a loss of generation with magnitude of 800MW(1000W in hardware) A smaller magnitude of disturbance in exper-imental evaluation compared to the reference imbalance magnitudeis due to the stability of the setup which put a limit on the sparepower capacity reserved for imbalance event As the objective ofpower hardware-in-the-loop evaluation is to demonstrate the capa-bility of the proposed control to operate under real measurementsthe magnitude of the disturbance does not impact the integrity ofthe study However it further demonstrates the applicability of theproposed solution to a wide range of disturbances The frequencymeasurements are derived from the raw voltage measurements atCell 2 and Cell 3 nodes using 3-phase voltage transformers A real-time system controller a multi-processor based system with analogto digital sampling and digital pre-filtering at 66666micros is utilizedto process the measurements As in the simulation based studiesthe measurement window to obtain RoCoF remains Tn = 002sThe parameters of the control system remain same as in the pre-vious study The system frequency response subject to generationloss at t = 10s is shown in Fig 7b As can be observed the ADDwas able to identify the imbalance evident from the restoration offrequency much faster than CLFC Therefore demonstration of thefeasibility of ADD operation with real measurements strongly sup-ports the deployment of the proposed control approach in futurepower systems

6 Conclusions

In this paper to improve the speed of response of secondary fre-quency control a novel control has been proposed The controlincorporates a fast acting balance control loop in addition to the sec-ondary control loop The balance control loop incorporates an areadistrubance determiner capable of fast autonomous and definitiveidentification of location of the imbalance The impact of fast anddefinitive identification of location of imbalance is two-fold

bull it ensures unilateral activation of reserves only within the areawhere the imbalance initiated This allows for faster response speedbull it enables use of area power imbalance observed over tie-lines ascontrol input as opposed to conventionally utilized area control errorThis avoids sympathetic activations even at faster response speeds

The proposed enhanced load frequency control further decouples thecontrol effort of fast acting resources from that of conventional gen-erators The control effort to fast acting resources is a combination ofeffort from balance control loop and secondary control loop whilethe control effort to conventional generators is from secondary con-trol loop only This allows for the exploitation of true effectiveness offast acting devices for frequency restoration The analytical assess-ments in the work undertaken provides a high degree of confidencein the aptness of the proposed approach for dynamically changingpower systems while the experimental evaluation demonstrates itsreal-world applicability

Acknowledgment

This work was supported by the European Commission under theFP7 ELECTRA IRP (grant no 609687) Any opinions findings andconclusions or recommendations expressed in this material are those

of the authors and do not necessarily reflect those of the EuropeanCommission

7 References1 lsquoNational Transmission Network Development Planrsquo (Australian Energy

Market Operator (AEMO) 2016) Available from httpwwwaemocomauElectricityNational-Electricity-Market-NEMPlanning-and-forecastingNational-Transmission-Network-Development-Plan

2 Torres LMA Lopes LAC MorAtildean TLA Espinoza CJR lsquoSelf-tuningvirtual synchronous machine A control strategy for energy storage systems to sup-port dynamic frequency controlrsquo IEEE Transactions on Energy Conversion 201429 (4) pp 833ndash840

3 Rakhshani E Remon D Cantarellas AM Rodriguez P lsquoAnalysis of deriva-tive control based virtual inertia in multi-area high-voltage direct current intercon-nected power systemsrsquo IET Generation Transmission Distribution 2016 10 (6)pp 1458ndash1469

4 Alipoor J Miura Y Ise T lsquoPower system stabilization using virtual syn-chronous generator with alternating moment of inertiarsquo IEEE Journal of Emergingand Selected Topics in Power Electronics 2015 3 (2) pp 451ndash458

5 Soni N Doolla S Chandorkar MC lsquoImprovement of transient response inmicrogrids using virtual inertiarsquo IEEE Transactions on Power Delivery 2013 28(3) pp 1830ndash1838

6 lsquoFast Frequency Response in the NEM Working Paperrsquo (Aus-tralian Energy Market Operator (AEMO) 2017) Available fromhttpswwwaemocomau-mediaFilesElectricityNEMSecurityandReliabilityReports2017FFR-Working-Paper---Finalpdf

7 lsquoSystem Operability Framework 2016rsquo (National Grid 2016) Availablefrom httpwww2nationalgridcomUKIndustry-informationFuture-of-EnergySystem-Operability-Framework

8 lsquoRoCoF - An independent analysis on the ability of Generators to ride throughRate of Change of Frequency values up to 2Hzsrsquo (DNV-KEMA for Eir-GridSONI 2013) Available from httpwwweirgridgroupcomsite-fileslibraryEirGridDNV-KEMA_Report_RoCoF_20130208final_pdf

9 lsquoInternational Review of Frequency Control Adaptationrsquo (DGA Consultingreport to Australian Energy Market Operator (AEMO) 2016) Available fromhttpswwwaemocomau-mediaFilesElectricityNEMSecurityandReliabilityReportsFPSS---International-Review-of-FrequencyControlpdf

10 lsquoPolicy P1 Load-Frequency Control and Performancersquo (in Continen-tal Europe Operation Handbook ENTSO-E 2016) Available fromhttpswwwentsoeeufileadminuser_upload_librarypublicationsentsoeOperation_HandbookPolicy_1_finalpdf

11 lsquoBalancing and Frequency Controlrsquo (North American Electric ReliabilityCorporation 2011) Available from httpwwwnerccomdocsocrsNERC20Balancing20and20Frequency20Control20040520111pdf

12 lsquoGuide to Ancillary Services in the National Electricity Mar-ketrsquo (Australian Energy Market Operator (AEMO) 2015)Available from httpswwwaemocomau-mediaFilesPDFGuide-to-Ancillary-Services-in-the-National-Electricity-Marketashx

13 Jin C Lu N Lu S Makarov YV Dougal RA lsquoA coordinating algorithm fordispatching regulation services between slow and fast power regulating resourcesrsquoIEEE Transactions on Smart Grid 2014 5 (2) pp 1043ndash1050

14 Guo Y Xie X Wang B Shi W Dong Y Mou L et al lsquoImproving agcperformance of a coal-fueled generators with mw-level bessrsquo In 2016 IEEE PowerEnergy Society Innovative Smart Grid Technologies Conference (ISGT) ( 2016pp 1ndash5

15 Xie X Guo Y Wang B Dong Y Mou L Xue F lsquoImproving agc perfor-mance of coal-fueled thermal generators using multi-mw scale bess A practicalapplicationrsquo IEEE Transactions on Smart Grid 2018 9 (3) pp 1769ndash1777

16 Zhang F Hu Z Xie X Zhang J Song Y lsquoAssessment of the effectivenessof energy storage resources in the frequency regulation of a single-area powersystemrsquo IEEE Transactions on Power Systems 2017 32 (5) pp 3373ndash3380

17 KintnerMeyer M lsquoRegulatory policy and markets for energy storage in northamericarsquo Proceedings of the IEEE 2014 102 (7) pp 1065ndash1072

18 lsquoFrequency Control by Demand Management (FCDM)rsquo (National Grid 2015)Available from httpwww2nationalgridcomWorkAreaDownloadAssetaspxid=41294

19 Short JA Infield DG Freris LL lsquoStabilization of grid frequency throughdynamic demand controlrsquo IEEE Transactions on Power Systems 2007 22 (3)pp 1284ndash1293

20 Hu J Cao J Guerrero JM Yong T Yu J lsquoImproving frequency stabilitybased on distributed control of multiple load aggregatorsrsquo IEEE Transactions onSmart Grid 2017 8 (4) pp 1553ndash1567

21 Variani MH Tomsovic K lsquoDistributed automatic generation control usingflatness-based approach for high penetration of wind generationrsquo IEEE Transac-tions on Power Systems 2013 28 (3) pp 3002ndash3009

22 Xie P Li Y Zhu L Shi D Duan X lsquoSupplementary automatic genera-tion control using controllable energy storage in electric vehicle battery swappingstationsrsquo IET Generation Transmission Distribution 2016 10 (4) pp 1107ndash1116

23 Keyhani A Chatterjee A lsquoAutomatic generation control structure for smartpower gridsrsquo IEEE Transactions on Smart Grid 2012 3 (3) pp 1310ndash1316

24 Kok K Widergren S lsquoA society of devices Integrating intelligent distributedresources with transactive energyrsquo IEEE Power and Energy Magazine 2016 14(3) pp 34ndash45

25 R LGraham Knuth DE Patashnik O lsquoConcrete Mathematics A Foun-dation for Computer Sciencersquo (Addison-Wesley Publishing Company1990) Available from httpswwwcsientuedutw~r97002tempConcrete20Mathematics202epdf

IET Research Journals pp 1ndash10ccopy The Institution of Engineering and Technology 2015 9

53 Controller and Power Hardware-in-the-Loop

The controller and power hardware-in-the-loop setup is shown inFig 7a As can be observed two areas of the GB network arerepresented by Cell 2 and Cell 3 of the Dynamic Power SystemsLaboratory [43] while the remaining three areas are simulated withinthe digital real-time simulator To ensure stability of the setup andto be able to represent a large portion of the GB network thecurrents are scaled by a factor of k = 08times 106 A detailed descrip-tion and process of initialization and synchronization of the powerhardware-in-the-loop setup is found in [44]

The system is subject to disturbance in LFC area 2 (Cell 2 inhardware Fig 7a) a loss of generation with magnitude of 800MW(1000W in hardware) A smaller magnitude of disturbance in exper-imental evaluation compared to the reference imbalance magnitudeis due to the stability of the setup which put a limit on the sparepower capacity reserved for imbalance event As the objective ofpower hardware-in-the-loop evaluation is to demonstrate the capa-bility of the proposed control to operate under real measurementsthe magnitude of the disturbance does not impact the integrity ofthe study However it further demonstrates the applicability of theproposed solution to a wide range of disturbances The frequencymeasurements are derived from the raw voltage measurements atCell 2 and Cell 3 nodes using 3-phase voltage transformers A real-time system controller a multi-processor based system with analogto digital sampling and digital pre-filtering at 66666micros is utilizedto process the measurements As in the simulation based studiesthe measurement window to obtain RoCoF remains Tn = 002sThe parameters of the control system remain same as in the pre-vious study The system frequency response subject to generationloss at t = 10s is shown in Fig 7b As can be observed the ADDwas able to identify the imbalance evident from the restoration offrequency much faster than CLFC Therefore demonstration of thefeasibility of ADD operation with real measurements strongly sup-ports the deployment of the proposed control approach in futurepower systems

6 Conclusions

In this paper to improve the speed of response of secondary fre-quency control a novel control has been proposed The controlincorporates a fast acting balance control loop in addition to the sec-ondary control loop The balance control loop incorporates an areadistrubance determiner capable of fast autonomous and definitiveidentification of location of the imbalance The impact of fast anddefinitive identification of location of imbalance is two-fold

bull it ensures unilateral activation of reserves only within the areawhere the imbalance initiated This allows for faster response speedbull it enables use of area power imbalance observed over tie-lines ascontrol input as opposed to conventionally utilized area control errorThis avoids sympathetic activations even at faster response speeds

The proposed enhanced load frequency control further decouples thecontrol effort of fast acting resources from that of conventional gen-erators The control effort to fast acting resources is a combination ofeffort from balance control loop and secondary control loop whilethe control effort to conventional generators is from secondary con-trol loop only This allows for the exploitation of true effectiveness offast acting devices for frequency restoration The analytical assess-ments in the work undertaken provides a high degree of confidencein the aptness of the proposed approach for dynamically changingpower systems while the experimental evaluation demonstrates itsreal-world applicability

Acknowledgment

This work was supported by the European Commission under theFP7 ELECTRA IRP (grant no 609687) Any opinions findings andconclusions or recommendations expressed in this material are those

of the authors and do not necessarily reflect those of the EuropeanCommission

7 References1 lsquoNational Transmission Network Development Planrsquo (Australian Energy

Market Operator (AEMO) 2016) Available from httpwwwaemocomauElectricityNational-Electricity-Market-NEMPlanning-and-forecastingNational-Transmission-Network-Development-Plan

2 Torres LMA Lopes LAC MorAtildean TLA Espinoza CJR lsquoSelf-tuningvirtual synchronous machine A control strategy for energy storage systems to sup-port dynamic frequency controlrsquo IEEE Transactions on Energy Conversion 201429 (4) pp 833ndash840

3 Rakhshani E Remon D Cantarellas AM Rodriguez P lsquoAnalysis of deriva-tive control based virtual inertia in multi-area high-voltage direct current intercon-nected power systemsrsquo IET Generation Transmission Distribution 2016 10 (6)pp 1458ndash1469

4 Alipoor J Miura Y Ise T lsquoPower system stabilization using virtual syn-chronous generator with alternating moment of inertiarsquo IEEE Journal of Emergingand Selected Topics in Power Electronics 2015 3 (2) pp 451ndash458

5 Soni N Doolla S Chandorkar MC lsquoImprovement of transient response inmicrogrids using virtual inertiarsquo IEEE Transactions on Power Delivery 2013 28(3) pp 1830ndash1838

6 lsquoFast Frequency Response in the NEM Working Paperrsquo (Aus-tralian Energy Market Operator (AEMO) 2017) Available fromhttpswwwaemocomau-mediaFilesElectricityNEMSecurityandReliabilityReports2017FFR-Working-Paper---Finalpdf

7 lsquoSystem Operability Framework 2016rsquo (National Grid 2016) Availablefrom httpwww2nationalgridcomUKIndustry-informationFuture-of-EnergySystem-Operability-Framework

8 lsquoRoCoF - An independent analysis on the ability of Generators to ride throughRate of Change of Frequency values up to 2Hzsrsquo (DNV-KEMA for Eir-GridSONI 2013) Available from httpwwweirgridgroupcomsite-fileslibraryEirGridDNV-KEMA_Report_RoCoF_20130208final_pdf

9 lsquoInternational Review of Frequency Control Adaptationrsquo (DGA Consultingreport to Australian Energy Market Operator (AEMO) 2016) Available fromhttpswwwaemocomau-mediaFilesElectricityNEMSecurityandReliabilityReportsFPSS---International-Review-of-FrequencyControlpdf

10 lsquoPolicy P1 Load-Frequency Control and Performancersquo (in Continen-tal Europe Operation Handbook ENTSO-E 2016) Available fromhttpswwwentsoeeufileadminuser_upload_librarypublicationsentsoeOperation_HandbookPolicy_1_finalpdf

11 lsquoBalancing and Frequency Controlrsquo (North American Electric ReliabilityCorporation 2011) Available from httpwwwnerccomdocsocrsNERC20Balancing20and20Frequency20Control20040520111pdf

12 lsquoGuide to Ancillary Services in the National Electricity Mar-ketrsquo (Australian Energy Market Operator (AEMO) 2015)Available from httpswwwaemocomau-mediaFilesPDFGuide-to-Ancillary-Services-in-the-National-Electricity-Marketashx

13 Jin C Lu N Lu S Makarov YV Dougal RA lsquoA coordinating algorithm fordispatching regulation services between slow and fast power regulating resourcesrsquoIEEE Transactions on Smart Grid 2014 5 (2) pp 1043ndash1050

14 Guo Y Xie X Wang B Shi W Dong Y Mou L et al lsquoImproving agcperformance of a coal-fueled generators with mw-level bessrsquo In 2016 IEEE PowerEnergy Society Innovative Smart Grid Technologies Conference (ISGT) ( 2016pp 1ndash5

15 Xie X Guo Y Wang B Dong Y Mou L Xue F lsquoImproving agc perfor-mance of coal-fueled thermal generators using multi-mw scale bess A practicalapplicationrsquo IEEE Transactions on Smart Grid 2018 9 (3) pp 1769ndash1777

16 Zhang F Hu Z Xie X Zhang J Song Y lsquoAssessment of the effectivenessof energy storage resources in the frequency regulation of a single-area powersystemrsquo IEEE Transactions on Power Systems 2017 32 (5) pp 3373ndash3380

17 KintnerMeyer M lsquoRegulatory policy and markets for energy storage in northamericarsquo Proceedings of the IEEE 2014 102 (7) pp 1065ndash1072

18 lsquoFrequency Control by Demand Management (FCDM)rsquo (National Grid 2015)Available from httpwww2nationalgridcomWorkAreaDownloadAssetaspxid=41294

19 Short JA Infield DG Freris LL lsquoStabilization of grid frequency throughdynamic demand controlrsquo IEEE Transactions on Power Systems 2007 22 (3)pp 1284ndash1293

20 Hu J Cao J Guerrero JM Yong T Yu J lsquoImproving frequency stabilitybased on distributed control of multiple load aggregatorsrsquo IEEE Transactions onSmart Grid 2017 8 (4) pp 1553ndash1567

21 Variani MH Tomsovic K lsquoDistributed automatic generation control usingflatness-based approach for high penetration of wind generationrsquo IEEE Transac-tions on Power Systems 2013 28 (3) pp 3002ndash3009

22 Xie P Li Y Zhu L Shi D Duan X lsquoSupplementary automatic genera-tion control using controllable energy storage in electric vehicle battery swappingstationsrsquo IET Generation Transmission Distribution 2016 10 (4) pp 1107ndash1116

23 Keyhani A Chatterjee A lsquoAutomatic generation control structure for smartpower gridsrsquo IEEE Transactions on Smart Grid 2012 3 (3) pp 1310ndash1316

24 Kok K Widergren S lsquoA society of devices Integrating intelligent distributedresources with transactive energyrsquo IEEE Power and Energy Magazine 2016 14(3) pp 34ndash45

25 R LGraham Knuth DE Patashnik O lsquoConcrete Mathematics A Foun-dation for Computer Sciencersquo (Addison-Wesley Publishing Company1990) Available from httpswwwcsientuedutw~r97002tempConcrete20Mathematics202epdf

IET Research Journals pp 1ndash10ccopy The Institution of Engineering and Technology 2015 9

26 Prostejovsky AM Marinelli M Rezkalla M Syed MH GuilloSansanoE lsquoTuningless load frequency control through active engagement of distributedresourcesrsquo IEEE Transactions on Power Systems 2018 33 (3) pp 2929ndash2939

27 lsquoElectricity Ten Year Statement 2016rsquo (National Grid 2016) Availablefrom httpwww2nationalgridcomUKIndustry-informationFuture-of-EnergyElectricity-ten-year-statement

28 Kundur P lsquoPower System Stability and Controlrsquo (McGraw-Hill Education1994) Available from httpsbooksgooglecoukbooksid=2cbvyf8Ly4AC

29 Xia J Dysko A lsquoUk transmission system modelling and validation for dynamicstudiesrsquo In IEEE PES ISGT Europe 2013 ( 2013 pp 1ndash5

30 Emhemed A Adam G Hong Q Burt G lsquoStudies of dynamic interactions inhybrid ac-dc grid under different fault conditions using real time digital simula-tionrsquo In 13th IET International Conference on AC and DC Power Transmission(ACDC 2017) ( 2017 pp 1ndash5

31 lsquoUCTE Operation Handbook Policy 1 Load-frequency control and performancersquo(European Network of Transmission System Operators for Electricity 2017)Available from httpswwwentsoeeupublicationssystem-operations-reportscontinental-europe-operation-handbook

32 lsquoUCTE Operation Handbook Appendix 1 Load-Frequency Control andPerformancersquo (European Network of Transmission System Operatorsfor Electricity 2017) Available from httpswwwentsoeeupublicationssystem-operations-reportscontinental-europe-operation-handbook

33 lsquoThe Power of Transformation Wind Sun and the Economics of Flexible PowerSystemsrsquo (International Energy Agency 2014) Available from httpswwwieaorgpublicationsfreepublicationspublicationThe_power_of_Transformationpdf

34 Zhang CK Jiang L Wu QH He Y Wu M lsquoDelay-dependent robust loadfrequency control for time delay power systemsrsquo IEEE Transactions on PowerSystems 2013 28 (3) pp 2192ndash2201

35 Ko KS Sung DK lsquoThe effect of cellular network-based communication delaysin an ev aggregatorrsquos domain on frequency regulation servicersquo IEEE Transactions

on Smart Grid 2017 pp 1ndash136 M JEAlam Muttaqi KM Sutanto D lsquoA novel approach for ramp-rate control

of solar pv using energy storage to mitigate output fluctuations caused by cloudpassingrsquo IEEE Transactions on Energy Conversion 2014 29 (2) pp 507ndash518

37 Lu N Zhang Y lsquoDesign considerations of a centralized load controller usingthermostatically controlled appliances for continuous regulation reservesrsquo IEEETransactions on Smart Grid 2013 4 (2) pp 914ndash921

38 lsquoNational Electricity Transmission System Performance Report 2015-2016 acircASReport to the Gas and Electricity Markets Authorityrsquo (National Grid 2016)Available from httpwww2nationalgridcomWorkAreaDownloadAssetaspxid=8589936830

39 P MAshton Saunders CS Taylor GA Carter AM Bradley ME lsquoIner-tia estimation of the gb power system using synchrophasor measurementsrsquo IEEETransactions on Power Systems 2015 30 (2) pp 701ndash709

40 Ramana NV lsquoPower System Operation and Controlrsquo (Pearson Education India2010) Available from httpsbooksgooglecoukbooksaboutPower_System_Operation_Controlhtmlid=Gox3x1ClV94Campredir_esc=y

41 Syed MH Burt GM Drsquohulst R Verbeeck J lsquoExperimental validation offlexibility provision by highly distributed demand portfoliorsquo In CIRED Workshop2016 ( 2016 pp 1ndash4

42 Dambrauskas P Syed MH Blair SM Irvine JM Abdulhadi IF BurtGM et al lsquoImpact of realistic communications for fast-acting demand sidemanagementrsquo CIRED - Open Access Proceedings Journal 2017 2017 (1)pp 1813ndash1817

43 lsquoDynamic Power Systems Laboratoryrsquo (University of Strathclyde Institute forEnergy and Environment 2018) Available from httpswwwulabequipmentcomfacilitydynamicpowersystems

44 GuilloSansano E Syed MH Roscoe AJ Burt GM lsquoInitialization and syn-chronization of power hardware-in-the-loop simulations A great britain networkcase studyrsquo Energies 2018 2018 11 (5) pp 1087

IET Research Journals pp 1ndash1010 ccopy The Institution of Engineering and Technology 2015