Probabilistic Transmission Planning:

Framework, Sample Analysis, & Tools

David MulcahyNorth Carolina State UniversityFREEDM Systems Center

1

Probabilistic Planning Motivation

Goal: Apply current tools to perform probabilistic transmission on Texas electricity system

- Evaluate modeling tools- Compare sample projects- Determine gaps and needs to be used

Focused on practical considerations of modeling and analysis

2

Outline

• Probabilistic Planning Introduction• Sample Case on ERCOT system

– Scenario Creation– Contingency Simulation– Risk-based indices

• Outstanding Modeling Issues

3

PROBABILISTIC PLANNING INTRODUCTION

4

Deterministic Planning

5

NERC TPL-001-4 outlines requirements for scenarios, contingencies, and allowable remedial actions

SensitivitiesLoad Cases

…

Peak

Off-Peak

New Gen

Trans Upgrade

Retire Gen

New Trans

…

G1 G2 Gn

….

T1 T2 Tn….

Contingencies

Simulate Scenarios

Any violation requires projects to mitigate issues

Scenario 1

Scenario 2 … Scenario

N

Compare Alternatives

Investment A > Investment

B

Compare cost, mitigation extent, and other benefits

5

Probabilistic Planning

6

G1 G2 Gn

0.10

….

T1 T2 Tn

….

Outage Statistics

Compare AlternativesSimulate Scenarios

Determine reliability impact such as loss of load

Scenario 1

Scenario 2 … Scenario

N

2.5Investment

A > Investment B

Include risk-based reliability indices

0.05 0.07

0.05 0.01 0.02

0 30

• Quantify risk of low probability but high impact scenarios• Additional decision criteria to compare transmission investments

Objectives:

Operating Point Distribution

6

Probabilistic Planning Framework

7

7

ERCOT CASE STUDY

8

ERCOT Case Study

• Electric Reliability Council of Texas (ERCOT) system model– Texas, USA– Independent System Operator

• Transmission > 100 kV

• Research performed at ERCOT in summer 2016

9

Statistical Input DataTransmission and Generation Outage Statistics

• Include samples of deep contingency events

(N-3, N-4, N-5, …)

• Future work should include probability common mode outages

Forecasted System Operating States

• Include wind outputs, distributed energy resources, electric vehicles, and other uncertain futures

• Not limited to bivariate distributions

10

10

Load Scenarios

• 2002-2013 historical data – Forecast for 2017

• Include load growth uncertainty• 481,800 hourly scenarios

11

7.9%

24.0%

36.0%

24.0%

7.9%

0.0%

10.0%

20.0%

30.0%

40.0%

-4% -2% 0% 2% 4%

Prob

abilit

y

Load Forecast Error

2017 Wind and Load Forecasts

12

Coincident ERCOT Wind Output and Load based on

2002-2013 weather

Load probability includes

uncertainty of forecasts

12

Stratified Sampling

13

Stratified Sample Space

Ensures analysis of

high impact but low

probability events

13

Strata Definitions

• From historical wind and load data including load growth forecast

14

Load Bounds (MW) Wind Bounds (MW)

Scenario Name Probability Number of

Hours Upper Lower Upper Lower

Load1\W1 2.97E-05 24 81,511 78,668 16,114 4,025

Load2\W1 2.69E-04 168 78,668 75,825 16,114 4,025

Load3\W1 1.25E-03 589 75,825 72,982 16,114 4,025

Load1\W2 3.13E-05 28 81,511 78,668 4,025 0

Load2\W2 3.67E-04 252 78,668 75,825 4,025 0

Load3\W2 1.84E-03 944 75,825 72,982 4,025 0

Total 3.78E-03 2,005 81,511 72,982 16,114 0

Contingency Statistics

• Generator outage data from Generator Availability Database System (GADS)

• Transmission outage data based on NERC Transmission Availability Database System (TADS) based on voltage class

15

Contingency Scenarios

• Enumerated ~8,000 N-2 (generator and transmission) contingencies in coastal region of ERCOT

• Analysis would ideally include higher level contingencies

• Statistics include frequency and mean time to repair to measure severity

16

Simulating Contingency Events

17

/Power Floww/ Remedial Actions

1. Re-dispatch generators

2. Adjust switched components

3. Adjust taps

N. Load Shedding

ContingencyList

/Contingency Impacts

FrequencyDurationSeverity

Power Flow Scenario

…Remedial actions solve

system violations resulting from contingency through

control actions

Load shedding give severity of events instead of binary

failures

17

Probabilistic Planning Tools

18

RBPSB(EPRI Research

Level Tool)

TransCARE(EPRI Research

Level Tool)

PSS/E(Siemens

Commercial Tool)

Self Developed code

(Post Processing)

18

Probabilistic Planning Tools

19

RBPSB(EPRI Research

Level Tool)

TransCARE(EPRI Research

Level Tool)

PSS/E(Siemens

Commercial Tool)

Self Developed code

(Post Processing)

19

Calculation of Probabilistic Indices

Quantifying risk indices

• Includes likelihood and severity of events

• Currently, no established criteria for acceptable risk or indices– System wide (e.g. expected unserved energy (EUE), frequency of loss of load,

etc.)

– Component or contingency level (e.g. probability and frequency of violations)

• Better for ranking alternatives than as absolute metric

20

20

Calculation of Probabilistic Indices

Potential uses of probabilistic criteria and indices

• Evaluate cost/benefit of investment alternatives

• Used to identify weak areas of system

• Benchmark reliability over time and define acceptable risk

21

21

Contingency Probability

• Individual element probabilities used calculate probability of scenario

• Assumes that deeper contingencies has at least the impact of similar higher contingency

• Deeper contingencies are dependent on shallower contingencies – i.e. Probability Gen A and Gen B out together is

subset of both probabilities of Gen A and Gen B

22

Expected Unserved Energy (EUE)

• Expected value of load shedding unmitigated by remedial actions

• PSSE was used to calculated EUE for each operational scenario (strata samples)

• Common list of contingencies for all scenarios

23

Indices Calculations

• EUE – Expected Unserved Energy

𝐸𝐸𝐸𝐸𝐸𝐸 = �𝑠𝑠∈𝑆𝑆

𝑃𝑃 𝑠𝑠 × 𝑎𝑎𝑎𝑎𝑎𝑎𝑖𝑖(𝐸𝐸𝐸𝐸𝐸𝐸𝑖𝑖,𝑠𝑠)

EUEi,s is unserved energy for operational scenario, i, in stratum, s

• IRI – Incremental Reliability Index

𝐼𝐼𝐼𝐼𝐼𝐼 =𝐸𝐸𝐸𝐸𝐸𝐸

𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃 𝐶𝐶𝑃𝑃𝑠𝑠𝑃𝑃

24

Modeling Assumptions

25

• Sampling from 4/6 high load strata

• 5 wind\load samples per strata

• ~8,000 enumerated N-2 contingencies for coast region

• Use ERCOT system in PSSE

• Base case without improvement is not N-1 secure

• Simulating 3 N-1 secure transmission options

• EUE assumes independence of failures and probability of wind\load strata

conditions

• Determining project rank based on comparison of EUE

25

Expected Unserved Energy

26

15.02

8.26

11.00 10.97

0

5

10

15

20

Base Project A Project B Project C

Expe

cted U

nser

ved E

nerg

y (M

Wh/y

r.)

EUE allows ranking by risk of losing load in terms of severity and frequency

Project A1 Project C2 Project B3

26

4 Strata Case

N/A

0.011

0.007

0.005

0.000

0.005

0.010

0.015

Base Project A Project B Project C

Incre

menta

l Reli

abilit

y Ind

ex

(MW

h/yr./$

Millio

n)

Incremental Reliability Index (IRI)

Capital Cost ($ Millions) - $ 590 $ 554 $ 806

27

IRI shows reliability improvement from

base case per investment cost

Project A1 Project B2 Project C3

4 Strata Case

174

95 86 109

-

40

80

120

160

200

Base Project A Project B Project C

EUE

(MW

h/yea

r)

Load3\W1 Load3\W2 Load2\W1 Load2\W2 Load1\W1 Load1\W2

EUE by Strata with 6 Strata28

Capital Cost ($ Millions) - $ 590 $ 554 $ 806

Contributions to Project EUE29

-

50

100

150

Avg.

EUE

by S

trata

(MW

h\yea

r)

Base Project A Project B Project C

-

50

100

Load1\W1 Load1\W2 Load2\W1 Load2\W2 Load3\W1 Load3\W2EUE

Contr

ibutio

n (M

Wh\y

ear)

Base Project A Project B Project C

OUTSTANDING MODELING ISSUES

30

Input Data Needs

• Improved element outage statistics

• Incorporate statistics of common mode outages– Common tower or substations– Cascading failures– Low probability but high impact

31

Sampling Needs

• Strata definitions?– Probability cutoffs– Operational conditions of interest

• Number of samples from strata?– What are the sufficient number of samples to

characterized the region?

• Sampling techniques and contingency depth?

32

Remedial Actions Needs

• Handling failed power flow solutions?– Can you ignore failures?– Are they indications of high impact events or

problem with numerical method?– Automate rerunning the cases?

• What are the appropriate remedial actions?

• Transparency of remedial action results?

33

Risk-Based Indices Needs

• Incorporating probability dependent deeper and shallower contingencies

• Comparison of different indices for decision support

• Determine how to be used for ranking alternatives or absolute metrics

34

Conclusions

• Barriers to application are primarily practical based on large, real systems

• Need improvements to modeling tools for remedial actions

• Useful as criteria to incorporate risk into current planning practices

35

David Mulcahydjmulcah@ncsu.edu

North Carolina State University

FREEDM Systems Center

Thank you!

36