· Web view2020/11/23 · নসর ল হ ম দ বল ন, সরক র নব য নয গ য জ ব ল ন র প রস র ন ন ভ ব প রণ দন দ
Smart Grid Topology Identification Using Sparse Recoverymsimoes/documents/papers2016/MGS_pape… ·...
Transcript of Smart Grid Topology Identification Using Sparse Recoverymsimoes/documents/papers2016/MGS_pape… ·...
0093-9994 (c) 2016 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TIA.2016.2574767, IEEETransactions on Industry Applications
Abstract -- Smart grid (SG) technology reshapes the traditional
power grid into a dynamical network with a layer of
information that flows along the energy system. Recorded data
from a variety of parameters in SGs can improve analysis of
different supervisory problems, but an important issue is their
cost and power efficiency in data analysis procedures. This
paper develops an efficient solution for power network topology
identification (PNTI) and monitoring activities in SG. The basic
idea combines optimization-based sparse recovery techniques
with a graph theory foundation. The power network (PN) is
modeled as a large interconnected graph, which can be
evaluated with the DC power-flow model. It has been shown that
topology identification for such a system can mathematically be
reformulated as a sparse recovery problem (SRP), and the
corresponding SRP can efficiently be solved using SRP solvers.
In this work we especially exploit the concentration of nonzero
elements in the corresponding sparse vectors around the main
diagonal in the nodal-admittance or structure matrix of the PN
to improve the results. The network models have been generated
with the MATPOWER toolbox, and Matlab-based simulation
results have indicated the promising performance of the
proposed method for real time TI in SGs.
Index Terms—Smart Grid, Sparse Recovery, Topology
Identification, Reweighted 𝒍𝟏-minimization.
I. INTRODUCTION
Smart grid (SG) technology reshapes the traditional power
grid from a single layer physical system to a huge dynamical
network that includes a second layer of information that
flows through the system [1]. This information layer is
formed from two major classes of data: the first class contains
the records collected from the status of different parameters
in the network such as bus voltages, powers, currents, and so
on [2], while the second class comprises the controlling
commands which are fed back to the network from decision
making units [13]. The supervisory control and data
acquisition (SCADA) system in addition to the technology of
the wide area monitoring system (WAMS) can provide such
massive voltage and power data in near real time [2]-[3].
Based on this vast amount of data, it is possible to define new
appropriate analysis tools which can transfer the standard
security and monitoring methods from static frameworks into
dynamical frameworks [12].
In addition to normal monitoring issues, PNTI and power
line status monitoring is particularly critical for a number of
1 M. Babakmehr, M.G. Simões, M.B. Wakin and F. Harirchi are with the
Division of Electrical Engineering, Department of Electrical Engineering and
Computer Science, Colorado School of Mines, Golden, CO 80401 USA, (e-mails: {mbabakme, msimoes, mwakin, harirchi}@ mines.edu). A. Al Durra
is with the Electrical Engineering Department, Petroleum Institute, Abu
Dhabi, (email: [email protected]). This work was partially supported by Petroleum Institute (PI) grant
470039 and NSF CAREER grant CCF-1149225.
tasks, including real-time contingency analysis, power-flow
analysis, power outage identification (POI), state estimation,
resiliency against natural disasters, and security assessment of
power systems [4]-[6]. Both network structural change and
fault identification also play a crucial role in power quality
improvement, system operation, and Microgrid technology
[7]. Smart Microgrid units will form the basis of the future
generation of power systems [8]. Numerous advanced
Microgrid design schemes have been developed to date [10],
where power network situational awareness is the most
critical point for islanding detection, system planning, and
protection [9], [11]. The system data exchange (SDX) module
of the North American Electric Reliability Corporation
(NERC) can provide grid-wide topology information on an
hourly basis. However, near real-time monitoring of
transmission lines is mandatory in order to make the PN act
as a smart system.
The PNTI problem has been addressed using economics in
[14]. Most other related works are focused on the problem of
fault or outage line detection in a power grid [4]-[5], [14]-
[21], [35]. In general, identifying the position of possible
outage lines can be interpreted as power network topology
identification; we call this the Power Outage Identification
Problem (POIP). Most of these works rely on the DC linear
power flow model (an approximation of the AC model) and
data measured from phasor measurement units (PMUs). The
POIP has been formulated as a combinatorial problem that is
computationally tractable only for single or double line
changes [4]. However, one must be able to cope with multiple
line changes in the face of cascading failures as in recent
blackouts. This has motivated several existing works [5],
[14]-[21]. A recent alternative approach for line-change
identification adopts a Gauss-Markov graphical model of the
power network and can deal with multiple changes [5];
Banerjee et al. [15] developed a new method for solving the
POIP based on the theory of quickest change detection, and
an ambiguity group-based location recognition algorithm has
been proposed in [16]. Moreover, a non-iterative method for
wide-area fault location has been presented in [17], which
applies the substitution theorem. In [18]-[20], the authors
used the sparse nature of the POIP to reformulate this
problem as a sparse recovery problem (SRP). In [21], a global
stochastic optimization technique based on cross-entropy
optimization (CEO) was presented. Most of these methods
require hourly base case grid topology information
(corresponding to the nodal-admittance matrix 𝐵, which we
will discuss) in addition to the system parameter data to be
able to find the outage lines. Also, a pre-whitening procedure
is mandatory for most of methods discussed above. There is a
rank deficiency issue when using such a grid topology matrix
[19]. In order to deal with this rank deficiency, one bus is
Smart Grid Topology Identification Using
Sparse Recovery Mohammad Babakmehr, Student Member, IEEE, Marcelo G. Simões, Fellow, IEEE,
Michael B. Wakin, Senior Member, IEEE, Ahmed Al Durra, Senior Member, IEEE,
Farnaz Harirchi, Student Member, IEEE1
0093-9994 (c) 2016 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TIA.2016.2574767, IEEETransactions on Industry Applications
normally considered as the reference bus, and its
corresponding row and column are removed from the matrix
𝐵 in order to obtain a full rank matrix [18]. However, due to
cyber-attacks and other precision deficiency issues, whenever
records from the reference bus are affected by bad data the
overall procedure might be completely corrupted. Another
important problem is with the switching events that are
mistakenly unreported. These kinds of errors can change the
structure of the nodal-admittance matrix 𝐵. Thus, if the last
status of the topology of the network has not been reported
within a suitable time interval, the final results can be
affected. In addition, to the best of our knowledge, none of
the methods mentioned above account for the random
behavior applied by renewable energy sources and the
uncertainty in loads [22].
Recently, we have addressed some of these issues in [23],
which employs a compressive sensing-based approach
involving some modifications to greedy-based sparse
recovery solvers. In this work we further investigate the
optimization-based formulations of the sparse PNTI, and we
aim to exploit additional structure within the sparse vectors in
order to improve the performance of the optimization-based
sparse recovery techniques for PNTI using a few
measurements of the system parameters (that can also be used
to solve the POIP). We particularly exploit and emphasize the
concentration of the nonzero elements in the corresponding
sparse vectors around the main diagonal of the nodal-
admittance or structure matrix of the PN, and we improve the
final identification performance using reweighted 𝑙1-
minimization [26].
Recently, sparse recovery has found interest in fault-type
identification and localization in distribution networks as well
[24], [25]. The approaches utilized here and in [23] are
completely different from the literature in the formulation.
Our methods rely only on the measurements from system
parameters and do not require any a priori information about
the topology of the network. In fact, we reformulate the PNTI
problem in such a way that the output of the optimization
problem is the structure or topology matrix 𝐵 of the network
itself. Case studies using IEEE standard test-beds [27] show
that the proposed method represents a promising fast and
accurate strategy for line change, fault detection, and
monitoring issues in SGs. This paper is the journal version of
our presented work in 2015 IAS annual meeting [36].
II. POWER NETWORK MODELING
This section describes the graph representation of power
networks, which are modeled with DC power flow equations.
In addition, we relate the Laplacian matrix of the graph to the
nodal-admittance matrix of the PN.
A. Network Model
In this work, the PN is modeled as a graph 𝐺(𝑆𝑁 , 𝑆𝐿),
consisting of a set of 𝑁 nodes 𝑆𝑁 = {1, . . . , 𝑁}, where each
bus of the PN is associated with a node 𝑖 ∈ 𝑆𝑁, and a set of 𝐿
edges or links 𝑆𝐿 ⊆ {𝑙𝑖,𝑗: 𝑖, 𝑗𝜖𝑆𝑁}, where each edge models a
transmission line.2 Following the same approach as [5] and
[18], the data from the buses’ electrical parameters such as
powers and voltages are assumed to be obtained using a
network of sensors. As an example, Fig.1 shows the structure
of the corresponding graph of the IEEE Standard-30 Bus.
Fig.1. Graph of IEEE Standard-30 Bus
B. DC Power Flow Model and its Graph
A simplified linearized approximation of the AC power
flow model has been introduced in [28] and [29]. This is
widely known as a DC load-flow (power-flow) model. Under
the DC load-flow regime, the active power injected to a
particular bus 𝑖 follows the superposition law
𝑃𝑖 = ∑ 𝑃𝑖𝑗𝑗 = ∑ 𝑏𝑖𝑗𝑗𝜖ℵ𝑖
(𝜃𝑖 − 𝜃𝑗) , (1)
where 𝑃𝑖𝑗 is the active power injected from node 𝑗 to node 𝑖,
𝜃𝑖 represents the phase of the voltage on each node, and 𝑏𝑖𝑗 is
the imaginary part of the admittance of line 𝑙𝑖,𝑗 or the
susceptance. Susceptance (B) is the imaginary part of
admittance (the inverse of impedance), in SI units,
susceptance is measured in Siemens. Considering, ℵ𝑖 as the
set of neighbor buses connected directly to bus 𝑖. It is useful
to rewrite these summations in a matrix-vector format, where
we have
𝒑 = 𝐵𝜽 𝒑, 𝜽 𝜖𝑅𝑁. (2)
In this equation the voltage phasor angle values of the nodes
are collected in the vector 𝜽𝜖𝑅𝑁, and the active power values
of the nodes are stored in the vector 𝒑𝜖𝑅𝑁. The matrix 𝐵 ∈𝑅𝑁×𝑁 is called the nodal-admittance matrix describing a
2 We will use bus/node and edge/(transmission) line interchangeably
within the rest of the manuscript.
Node 1
Node 2Node 3
Node 4 Node 5
Node 6
Node 7Node 8Node 9
Node 10 Node 11
Node 12
Node 13 Node 14
Node 15 Node 16
Node 17Node 18
Node 19
Node 20
Node 21
Node 22Node 23
Node 24
Node 25
Node 26 Node 27
Node 28
Node 29
Node 30
0093-9994 (c) 2016 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TIA.2016.2574767, IEEETransactions on Industry Applications
power network of 𝑁 buses, and it can be represented in the
following format:
𝐵𝑖𝑗 = {
−𝑏𝑖𝑗 , 𝑖𝑓 𝑙(𝑖,𝑗)𝜖𝑆𝐿
∑ 𝑏𝑖𝑗 ,𝑗𝜖𝛮𝑖 𝑖𝑓 𝑖 = 𝑗
0, 𝑜𝑡ℎ𝑒𝑟𝑤𝑖𝑠𝑒
} . (3)
It has been shown that this matrix can be viewed as a
weighted version of the Laplacian matrix of the graph
𝐺(𝑆𝑁 , 𝑆𝐿) [28], [18]. Since the nodal-admittance matrix gives
a full description of the structure of the networks’ graph, we
can represent the PNTI problem as that of determining the
structure of the nodal-admittance matrix 𝐵 [23]. As has been
discussed before [5], [22], and [30], power flow injection
originates from the aggregated load requests of a large
number of users and can be well approximated using
Gaussian random variables; in addition, there is uncertainty
caused by the utilization of renewable resources. In light of
this, the difference of phasor angles (𝜃𝑖(𝑡) − 𝜃𝑗(𝑡)) across a
bus in equation (2) can be approximated by a Gaussian
random variable at each sample of time [5], [22]. This
random behavior affects the structure of the sensing matrix
that is introduced in the next section; the randomness is
beneficial to the performance of sparse recovery techniques.
III. SMART GRID SPARSE TOPOLOGY IDENTIFICATION
We frame the PNTI problem as that of recovering the
topology matrix 𝐵 from the set of measurements of 𝑝 and 𝜃.
Similar to the CS-based approach that we have developed in
[23], our key assumption is that the PN is a sparse
interconnected system. This sparse structure helps us to
reformulate the PNTI problem as a sparse recovery problem
that can be solved using a small set of measurements in a fast
and accurate way using SRP solvers. This assumption has
been made based on our observations on a survey of standard
power grids models found on databases and toolboxes such as
MATPOWER. Considering a variety of standard PN models,
it can be observed that the regular maximum connectivity
level of an electrical bus in a grid is typically less than 5-
10%, especially in case of large-scale PNs. When the sparsity
assumption is not valid, however, the effectiveness of
sparsity-based PNTI will be limited.
A. PNTI Sparse Optimization-based Formulation
Given an interconnected power network of 𝑁 nodes or
buses, let the measurements of the active power and phase
angle of node 𝑖 be associated with the two following time
series, each of 𝑀 sample times:
𝑃𝑖(𝑡) , 𝜃𝑖(𝑡) 𝑓𝑜𝑟 𝑡 = 1,2, . . . , 𝑀 . (4)
Under the DC power flow regime, for each node in the
network and at each sample time 𝑡, we have the following
superposition law:
𝑃𝑖(𝑡) = ∑ 𝑃𝑖𝑗𝑗 (𝑡) = ∑ 𝑏𝑖,𝑗𝑗𝜖ℵ𝑖
(𝜃𝑖(𝑡) − 𝜃𝑗(𝑡)) . (5)
Here ℵ𝑖 is the set of neighbor buses connected directly to bus
𝑖 and 𝑏𝑖,𝑗 is the susceptance along the line 𝑙𝑖,𝑗 under the DC
model. Since 𝑏𝑖,𝑗 = 0 for 𝑗 ∉ ℵ𝑖, we can extend (5) as
follows:
𝑃𝑖(𝑡) = ∑ 𝑃𝑖𝑗
𝑗
(𝑡) =
∑ 𝑏𝑖,𝑗𝑗𝜖𝑆𝑁𝑖 (𝜃𝑖(𝑡) − 𝜃𝑗(𝑡)) + 𝑢𝑖(𝑡) + 𝑒𝑖(𝑡), (6)
where 𝑆𝑁𝑖 is the set of all nodes in the network except node 𝑖,
𝑢𝑖 is the possible leakage active power in node 𝑖 itself, and 𝑒𝑖
is the measurement noise. Representing the vector of active
power values as 𝒚𝒊, summing 𝒖𝒊 and 𝒆𝒊𝜖𝑅𝑀 (which we
assume to be modelled as a vectors of white Gaussian noise),
and dropping the time-sample notation, we end up with the
following equation for each node:
𝑦𝑖 = 𝐴𝑖𝑥𝑖 + 𝜂𝑖 (7)
𝐴𝑖 = [𝑎1,𝑖 , … , 𝑎𝑖−1,𝑖 , 𝑎𝑖+1,𝑖, . . . , 𝑎𝑁,𝑖] 𝜖 𝑅𝑀×𝑁−1 (8)
𝑎𝑗,𝑖(𝑡) = (𝜃𝑖(𝑡) − 𝜃𝑗(𝑡)) 𝑓𝑜𝑟 𝑡 = 1,2, . . . , 𝑀 (9)
𝑥𝑖 = [𝑏𝑖,1, … , 𝑏𝑖,𝑖−1, 𝑏𝑖,𝑖+1, . . . , 𝑏𝑖,𝑁]𝑇
𝜖𝑅𝑁−1, (10)
where 𝒚𝒊𝜖𝑅𝑀, and where 𝜼𝒊𝜖𝑅𝑀 is a vector of white
Gaussian noise. In our formulation, each column of the
matrix 𝐴𝑖 represents the difference between phase angles of
node 𝑖 vs. each node 𝑗𝜖𝑆𝑁𝑖 in the network for 𝑀 samples of
time. Since 𝑏𝑖,𝑗 = 0 for 𝑗 ∉ ℵ𝑖, the vector 𝒙𝒊 is a 𝐾-sparse
vector (see Definition.1) where 𝐾 is the number of nodes
which are directly connected to node 𝑖.3
Given 𝒚𝒊, solving for 𝒙𝒊 can be approached using a Least
Squares formulation. However, a large number of
measurements 𝑀 will be needed to solve such a problem in
the case of large power grids. Since all vectors 𝒙𝑖 are sparse
vectors, in order to reduce the measurement requirement, we
suggest using sparse recovery techniques [31]. Solving this
problem for each sparse vector 𝒙𝒊, we can concatenate all of
the sparse vectors together, form the nodal-admittance matrix
𝐵, and the process is completed4 (PNTI-SRP).
For each step the goal is to recover one individual column 𝑖 (termed 𝒙𝒊 in our formulation) of the nodal-admittance
matrix5 𝐵. Once this column is recognized, the connectivity
structure between bus 𝑖 and all other buses in the network can
be determined by considering the location of nonzero
elements in that specific column. In particular, if 𝑥𝑖(𝑗) has a
nonzero value we consider bus 𝑖 to be connected to bus 𝑗
through a transmission line with susceptance 𝑥𝑖(𝑗). If the
3 Therefore, the topology identification problem can be viewed as the
estimation of all sparse vectors {𝑥𝑖}𝑖=1𝑁 that best match the observed
measurements {𝑦𝑖 , 𝐴𝑖}𝑖=1𝑁 .
4 Since 𝜃𝑖(𝑡) − 𝜃𝑗(𝑡) = 0 for 𝑖 = 𝑗 & 𝑓𝑜𝑟 𝑡 = 1,2, . . . , 𝑀, we should
keep the node 𝑖 out of 𝑆𝑁𝑖 to avoid producing a vector of zeros in the
corresponding column of the matrix 𝐴𝑖. This means that we are not able to
find the value of the parameter 𝐵𝑖𝑖 directly from recovered vector 𝑥𝑖;
however, according to the definition of the nodal-admittance matrix 𝐵, 𝐵𝑖𝑖 =∑ 𝑏𝑖,𝑗𝑗𝜖𝛮𝑖
. After recovering the vector 𝑥𝑖, 𝐵𝑖𝑖 can be easily calculated.
5 Regarding the mathematical formulation, the recovery step can be done individually and in parallel over all columns.
0093-9994 (c) 2016 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TIA.2016.2574767, IEEETransactions on Industry Applications
estimated 𝑥𝑖(𝑗) does not equal the estimated 𝑥𝑗(𝑖) for some
pair 𝑖 and 𝑗, this could be a sign of an outage or an anomaly
warranting further investigation.
In general, standard techniques for solving SRPs can be
categorized into two major groups: (𝑖) greedy algorithms, and
(𝑖𝑖) convex optimization based algorithms [16]. In [23] we
have applied the OMP greedy algorithm as a basic SRP
solver. We also presented modifications to the OMP
algorithm to handle certain practical problems such as a data
correlation issue. In this work we aim to discuss the
optimization based-rather than greedy-formulations of the
sparse PNTI, and in particular we exploit and emphasize the
concentration of nonzero elements in the corresponding
sparse vectors around the main diagonal of the nodal-
admittance matrix 𝐵. To exploit this structure, we present a
new PNTI formulation using reweighted 𝑙1-minimization.
B. 𝑆𝑝𝑎𝑟𝑠𝑒 𝑅𝑒𝑐𝑜𝑣𝑒𝑟𝑦 𝐹𝑢𝑛𝑑𝑎𝑚𝑒𝑛𝑡𝑎𝑙𝑠
The sparse recovery problem (SRP) can be interpreted as a
paradigm for recovering an unknown signal from a set of
underdetermined linear measurements. The ability to solve an
SRP requires the assumption of sparsity for the signal to be
recovered, and it requires the measurement or sensing matrix
to satisfy certain conditions.6
Definition 1: A 𝐾-sparse signal 𝒙𝜖𝑅𝑁 is a signal of length
𝑁 with 𝐾 nonzero entries where 𝐾 < 𝑁 (in many cases
(𝐾 << 𝑁)). The sparsity level of a signal 𝑥 is denoted by the
𝑙0 norm ‖𝑥‖0.
Intuitively, an SRP is an optimization problem in which the
goal is to recover a 𝐾-sparse signal 𝒙𝜖𝑅𝑁 from a set of
observations 𝒚 = 𝐴𝒙 𝜖𝑅𝑀 where 𝐴𝜖𝑅𝑀×𝑁 is the sensing
matrix with 𝑀 < 𝑁 (in many cases 𝑀 << 𝑁) [31]-[32]. Due
to the underdetermined nature of this recovery problem (since
𝑀 < 𝑁), the null space of the matrix 𝐴 is non-trivial;
therefore, there exist infinitely many candidate solutions for
this problem. However, under certain conditions on the
sensing matrix 𝐴, various sparsity based recovery methods
can be guaranteed to efficiently find the candidate solution
that is sufficiently sparse. Examples of such conditions are
the restricted isometry property (RIP), the exact recovery
condition, and low coherence. In essence, what these
conditions require is that any two small subsets of columns of
the sensing matrix 𝐴 must be almost orthogonal to each other.
Definition 2: An 𝑀 by 𝑁 sensing matrix 𝐴 is said to satisfy
the Restricted Isometry Property (RIP) of order 𝐾 if there
exists a constant 𝛿𝐾𝜖(0,1) such that:
(1 − 𝛿𝑘)‖𝑥‖22 ≤ ‖𝐴𝑥‖2
2 ≤ (1 + 𝛿𝑘)‖𝑥‖22 (11)
holds for all 𝐾-sparse vectors 𝒙. The parameter 𝛿𝐾 is known
as the isometry constant of order 𝐾.
Definition.3: The coherence of an 𝑀 × 𝑁 matrix 𝐴 is the
maximum normalized inner product (correlation) between
any two distinct columns of 𝐴:
6 The notation in this Section matches that in Section III-A, except for
convenience we suppose the unknown vector 𝑥 has length 𝑁 rather than 𝑁 − 1.
𝜇A = max1≤𝑚,𝑛≤𝑁,𝑚≠𝑛
|⟨𝑎𝑚,𝑎𝑛⟩|
‖𝑎𝑚‖2‖𝑎𝑛‖2. (12)
The following sparse PNTI formulations have been
developed based on sparse recovery theorems [31]-[32].
C. 𝑙1-Minimization Based PNTI
If the 𝑀 by 𝑁 sensing matrix 𝐴 satisfies the RIP of order
2𝐾 with any isometry constant less than 1, then the following
optimization problem can recover the original 𝐾-sparse signal
𝒙 from the set of measurements 𝒚 = 𝐴𝒙:
𝑃0: �̂� = 𝑎𝑟𝑔𝑚𝑖𝑛�́� ‖�́�‖0 𝑠𝑢𝑏𝑗𝑒𝑐𝑡 𝑡𝑜 𝑦 = 𝐴𝑥 ́ . (13)
In general, this 𝑙0-minimization problem is known to be
NP-hard. Fortunately, there is a relaxed version of this 𝑙0-
minimization problem, known as 𝑙1-minimization, that can
still guarantee the recovery of the sparse signal
𝑃1: �̂� = 𝑎𝑟𝑔𝑚𝑖𝑛�́�
‖�́�‖1 𝑠𝑢𝑏𝑗𝑒𝑐𝑡 𝑡𝑜 𝑦 = 𝐴𝑥 ́ . (14)
Here, ‖�́�‖1 = ∑ |�́�(𝑛)|𝑁𝑛=1 . Since the 𝑙1-norm is convex, this
results in a tractable convex optimization problem, widely
known as Basis Pursuit (BP).
D. Noisy 𝑙1-Minimization Based PNTI
In the case of noisy measurements, one can solve the
following problem instead:
𝑁𝑃1: �̂� = 𝑎𝑟𝑔𝑚𝑖𝑛�́�
‖�́�‖1 𝑠𝑢𝑏𝑗𝑒𝑐𝑡 𝑡𝑜 ‖𝑦 − 𝐴�́�‖2 < 𝜂. (15)
This problem is widely known as Basis Pursuit De-Noising
(BPDN). Suppose 𝐴 satisfies the RIP condition of order 2𝐾
with isometry constant 𝛿2𝐾 < 0.4651. Let 𝒚 = 𝐴𝒙 + 𝒏 be
noisy measurements of any vector x. If 𝜂 ≥ ‖𝑛‖2, then the
solution 𝒙 to (15) obeys:
‖𝑥 − �̂�‖2 ≤ 𝐶1‖𝑥−𝑥𝑆‖1
√𝑆+ 𝐶2𝜂, (16)
where 𝐶1 and 𝐶2 depend only on 𝛿2𝐾.7
In general, it is difficult to check whether a given sensing
matrix 𝐴 satisfies the RIP condition (11). An alternative, but
weaker property known as coherence (12) is easier to check
in practice. For a 𝐾-sparse vector 𝒙 and a vector of
measurements 𝒚 = 𝐴𝒙, if 𝜇𝐴 <1
2𝐾−1 , then, 𝑃1 can recover
the original 𝐾-sparse vector 𝒙 from the set of measurements
𝒚. Fig.2 shows how the average value of the coherence of the
sensing matrices 𝐴𝑖 (for 𝑖 = 1: 30) of IEEE Standard-30 Bus
changes as the number of measurements 𝑀 increases. The
smaller the coherence, the larger the permitted value of 𝐾,
and the broader the class of nodal-admittance matrix columns
(sparse vectors) 𝒙𝒊 that can be recovered. The minimum
number of measurements needed for perfect recovery is
related to both the original dimension 𝑁 of the signal 𝒙
(which in sparse PNTI equals the number of buses of the PN)
and the sparsity level 𝐾 (which equals the maximum in-
degree of buses); specifically, 𝑀 must be at least proportional
7 The vector 𝑥𝐾 is the closest 𝐾-sparse approximation to 𝑦
0093-9994 (c) 2016 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TIA.2016.2574767, IEEETransactions on Industry Applications
to 𝐾 𝑙𝑜𝑔(𝑁/𝐾). Successful recovery with 𝐾 𝑙𝑜𝑔(𝑁/𝐾) is
generally possible (with high probability) when the sensing
matrix 𝐴 is generated randomly, such as with independent
and identically distributed (i.i.d.) Gaussian entries.
E. Reweighted 𝑙1-Minimization Based PNTI
Some specialized methods have also been introduced
which are able to exploit extra prior information to solve the
SRP more efficiently and more accurately from even fewer
measurements than 𝑃1 or 𝑁𝑃1. Examining the structure of the
columns of the Laplacian (nodal-admittance) and adjacency
matrices for some standard IEEE test-beds (Fig.3) we can see
that due to the roughly sequential numbering of neighbor
buses within a PN, the positions of the nonzero elements in
both of these matrices exhibit a certain structure. In
particular, Fig.3 shows that most of the nonzero entries
concentrate close to the main diagonal. This means that in
each sparse vector 𝒙𝒊, certain entries are more likely than
others to be nonzero.
Fig.2. Average coherence versus number of measurements over all
corresponding sensing matrices 𝐴𝑖 of all of the nodes in IEEE Standard-30 Bus system, where the curve is averaged over 100 realizations of the network
Fig.3. Corresponding normalized Laplacian matrices of (a) IEEE Standard-57 Bus and (b) IEEE Standard-118 Bus, and the adjacency matrices of
(c) IEEE Standard-57 Bus and (d) IEEE Standard-118 Bus.
To exploit this anticipated behavior, we can replace 𝑃1
with the following weighted optimization problem instead:
𝑊𝑃1: �̂� = 𝑚𝑖𝑛�́�
∑ 𝜔(𝑛)|�́�(𝑛)|𝑁𝑛=1 𝑠𝑢𝑏𝑗𝑒𝑐𝑡 𝑡𝑜 𝑦 = 𝐴𝑥 ́, (17)
where 𝜔(1), 𝜔(2), … , 𝜔(𝑁) are non-negative weights. Similar
to 𝑃1, 𝑊𝑃1 is a convex optimization problem that can be
solved as a linear program (LP). By assigning smaller
weights near the diagonal elements, we can encourage those
elements to be selected in the recovered sparse signal 𝒙𝑖. 𝑊𝑃1
can be solved within a single step using our prior knowledge
for setting the weights. However, in order to increase the
accuracy of the final results we have extended this
formulation through an iterative procedure called Reweighted
𝑙1 minimization. This iterative algorithm (𝑅𝑤𝑙1) updates the
weights in each step based on the estimated sparse vector
magnitudes from the previous step (see Algorithm.1). In
Algorithm.1, 휀 is defined as a stabilizer parameter that is
used in order to obviate the effect of zero-valued components
in 𝒙𝑙. It has been shown that in general the 𝑅𝑤𝑙1 recovery
process tends to be reasonably robust to the choice of this
parameter (휀 > 10−3 is suggested for practical situations).
The diagonal matrix 𝑊𝑙 is defined as follows:
𝑊𝑙 = [
𝜔(1)𝑙 ⋯ 0
0 ⋱ 0
0 ⋯ 𝜔(𝑁)𝑙
]. (18)
Algorithm.1 Reweighted 𝒍𝟏-minimization (𝑹𝒘𝒍𝟏) PNTI
require: stopping criterion, phase angle measurements,
active power measurements, each of 𝑴 sample times
𝒑𝒊(𝒕) , 𝜽𝒊(𝒕) 𝒇𝒐𝒓 𝒕 = 𝟏, 𝟐, . . . , 𝑴.
1. form: measurement matrix 𝑨 ∈ 𝑹𝑵−𝟏×𝑴, and
measurement vector 𝒚
2. Set counter 𝒍 to zero and initialize 𝝎(𝒏)𝟎 𝒇𝒐𝒓 𝒏 = 𝟏: 𝑵 − 𝟏
3. solve:
𝒙𝒍 = 𝒂𝒓𝒈𝒎𝒊𝒏‖𝑾(𝒍)𝒙‖𝟏
𝒔𝒖𝒃𝒋𝒆𝒄𝒕 𝒕𝒐 𝒚 = 𝑨𝒙
4. update weights:
𝝎(𝒏)𝒍+𝟏 =
𝟏
|𝒙(𝒏)𝒍 | + 𝜺
𝒇𝒐𝒓 𝒏 = 𝟏: 𝑵 − 𝟏
5. Go back to step.3. Until stopping criterion met.
6. output: 𝒙 = 𝒙𝒍
F. Noisy Reweighted 𝑙1-Minimization Based PNTI
In short, to adapt the 𝑅𝑤𝑙1 algorithm in the case of noisy
measurements the following change should be made in the 3rd
step:
𝑥𝑙 = 𝑎𝑟𝑔𝑚𝑖𝑛‖𝑊(𝑙)𝑥‖1
𝑠𝑢𝑏𝑗𝑒𝑐𝑡 𝑡𝑜 ‖𝑦 − 𝐴𝑥‖𝑙2< 𝜂. (19)
Additional discussion of 𝑊𝑃1 and 𝑅𝑤𝑙1 is contained in [26].
There are a variety of optimization packages that can be used
to solve 𝑃1, 𝑁𝑃1, 𝑊𝑃1 and 𝑅𝑤𝑙1; examples include CVX [33]
and NESTA [34].
(b)
(c)
(d)
(a)
0093-9994 (c) 2016 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TIA.2016.2574767, IEEETransactions on Industry Applications
Our justification for this modeling assumption of quasi-
sequential node numbering was based on the structure
observed in IEEE standard test beds as well as some real-
world power networks. We believe that this modeling
assumption will be valid in a considerable number (but not
all) networks. Moreover, whenever a new area is constructed
and is added to a power network the new buses will usually
be numbered similarly to their neighbors, so some sequential
numbering may be preserved. In case such a sequential
numbering is not present within a network then one may see
only limited improvement from weighted 𝑙1 minimization.
Even in such a situation, one may still use ordinary sparsity-
based optimization approaches (including the iterative re-
weighted 𝑙1-minimization algorithm, but with weights
initialized to 1) to solve the problem, although with a slightly
greater number of measurements.
IV. SIMULATION RESULTS AND DISCUSSION
The authors tested the proposed method for recovering the
topology of a SG using compressive observations, collected
from the system parameters. In these simulations, we used the
IEEE Standard-30 Bus and IEEE Standard-300 Bus as case
studies. These power networks include 30, 300, and 2383
buses and 47, 411, and 2896 power transmission lines, and
their detailed specifications have been fully described in
MATPOWER toolbox [27]. The MATPOWER toolbox is
used for solving the power flow equations in various
demands and the resulting phase angle and active power
measurements are applied as the input to the sparse solver. In
order to generate the data, first, we fed the system with
Gaussian demands and simulated the PN. Based on new PMU
technology standards, measurement SNR has been randomly
set to 20-100 dB.
Fig.4. Recovery rate comparison of nodes a) 3 (in-degree 2), b) 10 (in-degree
6) in the network of Fig. 1, respectively. Success rate is calculated over 100
realizations of the network for a given number of measurements.
Within the network graphs, SRP solvers have different
recovery performance for different nodes, mainly because of
the sparsity level of the signal (or in our PN, in-degree or the
number of incoming transmission lines to an individual bus).
Fig.4.a, and b show the recovery performance of the 𝐵𝑃 and
𝑅𝑤𝑙1 solutions for nodes 3 and 10 of the IEEE Standard-30
Bus, respectively. These buses are distinguished from each
other by their in-degrees. The success rate8 is calculated over
100 realizations of the network for a given number of
measurements. From the sparsity level viewpoint, the 6-
sparse signal 𝑥10 corresponds to one of the most complicated
signals 𝑥𝑖 to be recovered in the IEEE Standard-30 Bus. The
2-sparse signal 𝑥3 (corresponding to the double in-connection
node 3) is more likely to be recovered using a smaller number
of measurements than the 6-sparse signal 𝑥10 (corresponding
to the 6 in-connection node 10). In Fig.5, the same trends can
be observed by looking at the recovery rate over the 1st, 2nd,
15th, 3rd and 130th buses, with sparsity levels (in-degrees) 3, 4,
6, 7, and 9, respectively, in the IEEE Standard-300 Bus.
Fig.5. Recovery rate comparison of nodes (downward): 1 (in-degree 3), 2 (in-
degree 4), 15 (in-degree 6), 3 (in-degree 7), 130 (in-degree 9) respectively.
Success rate is calculated over 100 realizations of the IEEE Standard-300
Bus for a given number of measurements.
We occasionally observe that nodes with similar in-degree
actually have a different recovery performance. For example,
8 For our success criterion we first check to see whether the support (the
positions of the non-zeros) of the true signal is correctly identified, and if so,
we then check to see whether the recovery error ‖𝑥𝑖 − 𝑥�̂�‖2/‖𝑥𝑖‖2 is within a
certain bound 휀. Both conditions must be satisfied for recovery to be
considered successful. We picked 휀 = 0.05 to illustrate the acceptable recovered signals.
0093-9994 (c) 2016 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TIA.2016.2574767, IEEETransactions on Industry Applications
nodes 3 and 15 (corresponding to the 7 and 6-sparse signals
𝑥3 and 𝑥15, respectively) have close in-degrees, but Fig.5
shows that the recovery rate curve of 𝑥3 is closer to that of
the 9-sparse signal 𝑥130, i.e., it requires a larger number of
measurements for correct recovery. This difference can be
caused by the network-wide location of the bus and also the
structure of the incoming transmission lines to each bus.
Moreover, due to the close geographical position and
similarity in load pattern within an interconnected network,
the parameters of different nodes may share a level of
correlation. This will affect the coherence of the resulting
sensing matrix and may result in different recovery
performance over the network nodes. Figs.4-5 also show how
the presence of structured sparsity helps the 𝑅𝑤𝑙1 algorithm
to outperform 𝐵𝑃, especially in case of the IEEE Standard-
300 Bus where the in-degree of the nodes is larger. Figs.6a-c
provide a node-by-node9 comparison of the recovery
performance for the 2 aforementioned SRP approaches over
100 realizations of the IEEE Standard-30 Bus. The vertical
axis indicates the number of measurements 𝑀, while the
horizontal axis represents the bus number from 1 to 30.10 A
color spectrum ranging from dark blue (corresponding to 0%
recovery) up to dark red (corresponding to 100% recovery)
has been used to illustrate the recovery performance
percentage. Results indicate that, for almost all nodes, both of
the algorithms can arrive at full recovery performance using
20-25 measurements. In general, however, the 𝑅𝑤𝑙1
algorithm demonstrates better performance with lower
numbers of measurements, especially for columns with high
congestion around the diagonal elements. Finally, Fig.7a-c
demonstrate the network-wide topology recovery
performance for 𝐵𝑃 and 𝑅𝑤𝑙1 on the IEEE Standard-30, 300,
and 2383 Bus networks, over 100 realizations of each
network, respectively. For each curve, the vertical axis
represents the percentage of trials in which all 30, 300, or
2383 columns of the corresponding nodal-admittance matrix
𝐵 (and, as a result, the network-wide topology) are
successfully recovered.
As has been discussed, theoretically, the number of
measurements required for full recovery 𝑀 should be at least
proportional to 𝐾 𝑙𝑜𝑔(𝑁/𝐾). In the case of the IEEE
Standard-30 Bus, node 6 has the highest in-degree of 7; as a
result the whole network-wide recovery is expected at
roughly 𝑀 = 7 𝑙𝑜𝑔(30/7) ≈ 10 or more. Moreover, based
on the discussion in Section III-B, as the number of buses of
a particular network, 𝑁, grows, more measurements are
needed. For example, in case of the IEEE Standard-300 Bus,
node 276 has in-degree 12, and full recovery is theoretically
possible when roughly 𝑀 = 12 𝑙𝑜𝑔 (300
12) ≈ 39. In the case
of the IEEE Standard-2383 network, node 1920 has in-degree
9; as a result, full recovery should be expected at 𝑀 =
9 𝑙𝑜𝑔 (2383
9) ≈ 50 measurements or more. However, Figs.6-7
indicate that in practice the whole network topology for these
9 That is, for each of the columns of the nodal-admittance matrix 𝐵. 10 Or equivalently, the column number from the nodal-admittance matrix.
two systems can typically be recovered from 𝑀 ≈ 20, 60 or
90 measurements per bus, respectively. This is likely due to
the coherence of the sensing matrices (because the elements
are not perfectly independent Gaussian random variables) and
the additional noise.
An important factor is that, although the number of
measurements needed for full recovery increases with the
network size 𝑁, this is not a linear relationship. As can be
seen, although the IEEE Standard-2383 and 300 Bus are
almost 80 and 10 times larger than IEEE Standard-30 Bus in
scale, respectively, they require less than 3 and 5 times more
measurements for full recovery, respectively. This fact
highlights the suitability of the sparse PNTI setup especially
for large scale power networks.
Fig.6. Recovery performance for 2 SRP solvers over all of the buses of the
standard IEEE Standard-30 Bus: (a) 𝐵𝑃, (b) 𝑅𝑤𝑙1, (c) 𝑅𝑤𝑙1 − 𝐵𝑃 .
(a)
(b)
(c)
0093-9994 (c) 2016 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TIA.2016.2574767, IEEETransactions on Industry Applications
In general, a lack of measurements reduces the recovery
performance for any SRP solver; however, the reweighted 𝑙1
minimization algorithm (𝑅𝑤𝑙1), which uses additional
structural knowledge on the sparse vectors, is less affected
and exhibits a better recovery performance when compared
with 𝐵𝑃. Finally, we do note that it is not possible to recover
the nodes with higher in-degrees until 𝑀 is large enough that
the PN coherence metric reaches a suitably small level. If
such a level were never to be reached, we might be limited in
the degree of nodes that we could recover with this technique.
Fig.7. Whole network topology recovery performance for 2 SRP solvers
for a) the IEEE Standard-30 Bus, b) the IEEE Standard-300 Bus over 100
realizations of the network, and c) the standard IEEE Standard-2383 Bus
over 100 realizations of the network.
V. CONCLUSIONS
In this paper we described a novel approach to address
topology identification and line outage localization in smart
power grids, using graph and sparse recovery theories. We
have discussed the computational complexity and expensive
data analysis required for large scale PNs and the uncertainty
of system states and records from parameters that may have
caused by uncertain behavior of the load. In addition, cyber-
attacks may cause unreported status changes, and there may
be random behavior of distributed generation systems (such
as wind turbines or PV cells). We presented an efficient and
low-cost solution for topology identification of line changes
and/or fault detection, plus possible monitoring tasks. This
method can efficiently overcome the inherent challenges in
the analysis of big data (that can capture different parameters
in the network). The presented approach formalizes the use of
the theory of sparse recovery for enhancing the smart-grid
analysis. The PN has been modeled as a sparse
interconnected graph. Therefore, the topology identification
problem was mathematically reformulated as a sparse
recovery problem (PNTI-SRP) based on the application of a
DC power-flow model. The conclusion is the appropriate use
of SRP solvers can solve the TI problem for IEEE standard
networks. The network model used in this study was
generated with the MATPOWER toolbox through standard
IEEE 30, 300, and 2383 bus test-beds. Such a sparse
reformulation for the TI problem depends only on
measurements from the system parameters; there is no need
for previous information from the network topology. In the
technique presented in this paper, SRP solver methods
supported finding the structure or topology matrix of the
network. Several case studies demonstrated that our proposed
method represents a promising alternative strategy for
topology identification, fault detection, and monitoring issues
using only a small set of observations from some of the bus
parameters. The recovery performance of the SRP solvers is
mainly dependent on the in-degree (number of lines
connected to each bus) of each bus in the network. Moreover,
it has been shown that columns of the matrix 𝐵 of a sample
PN (each corresponds to one of the sparse vectors 𝒙𝒊 to be
recovered in the 𝐵𝑃 or 𝑅𝑤𝑙1 solutions), can be assumed as
structured-sparse signal. This structural assumption can be
used to improve the recovery performance, using an even
smaller number of measurements, which makes the retrofit
and real-time application very easy to be done with
inexpensive hardware.
Finally, we note that within the state of the art, most of the
proposed power line status identification (essentially, PNTI)
algorithms rely on the voltage phasor analysis within the DC
power flow model and try to define a mathematical
formulation that can directly extract the structure of the
nodal-admittance or another structural matrix or criterion of
the PN. Our algorithm also fits into this category; such data
has traditionally been the readily available. However, PMU
technology today also provides real time measurements from
electrical current phasors. This functionality could be utilized
in order to develop new current phasor-based frameworks
which directly identify the status of power lines from the
current data. However, due to the uncertainty in loads and
within the aggregation of renewables, in addition to
nonlinearity in demand within various scales of power
networks, the general behavior of electrical currents may vary
over a wide range. Accounting for this uncertainty may
require a deep and comprehensive study before current
measurements can be used as a trustable criterion for
topology identification and power outage detection.
(a)
(b)
(c)
0093-9994 (c) 2016 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TIA.2016.2574767, IEEETransactions on Industry Applications
REFERENCES
[1] X.Y.Yu, C. Cecati, T. Dillon, and M. G. Simões, “The new frontier of
smart grids,” IEEE Ind. Electron. Mag., vol. 5, no. 3, pp. 49–63, Sep. 2011. [2] X. Fang, S. Misra, G. Xue, and D. Yang, “Smart grid—The new and
improved power grid: A survey,” IEEE Commun. Surveys Tutorials, vol. 14,
no. 4, pp. 944–980, Fourth Quarter 2012. [3] J. De La Ree, V. Centeno, J. S. Thorp, and A. G. Phadke. Synchronized
phasor measurement applications in power systems. IEEE Trans. Smart
Grid, 1(1):20–27, 2010. [4] J. E. Tate and T. J. Overbye, “Line outage detection using phasor angle
measurements,” IEEE Trans on Power Syst, vol. 23, pp: 164-165, Nov. 2008.
[5] M. He and J. Zhang. A dependency graph approach for fault detection and localization towards secure smart grid. IEEE Trans on Smart Grid, vol:2,
pp:342–351, June 2011.
[6] B. Ansari and S. Mohagheghi. "Optimal energy dispatch of the power distribution network during the course of a progressing wildfire". Int. Trans.
Electr. Energ. Syst. 2015; 25:3422–3438.
[7] A.S. Bubshait, M.G. Simoes, A. Mortezaei, and T.D.C. Busarello. "Power quality achievement using grid connected converter of wind turbine
system." In Proc. IEEE IAS Annu. Meeting, Oct. 2015, pp.1–8., Dallas, USA.
[8] S. Thale, R. Wandhare, V Agarwal, "A Novel Reconfigurable Microgrid Architecture with Renewable Energy Sources and Storage", IEEE
Transactions on Industry Appl. Vol. 51, No. 2, pp: 1805:1816, March 2015.
[9] F. Harirchi, M.G. Simoes, A. Al-Durra, S.M.Muyeen and M. Babakmehr. “Designing Smart Inverter with Unified Controller and Smooth Transition
between Grid Connected and Islanding Modes for Microgrid Application”, In
Proc. IEEE IAS Annu. Meeting, Oct. 2015, pp. 1–8, Dallas, USA. [10] M.G. Simoes, T.D.C. Busarello, A.S. Bubshait, F. Harichi, J.A. Pomilio,
and F. Blaabjerg, “Interactive smart battery storage for a PV and wind hybrid
energy management control based on conservative power theory,” International Journal of Control, in press, DOI: 10.1080/00207179.
2015.1102971, 2016.
[11] A. Luna, J. Rocabert, J. Candela, J. Hermoso, R. Teodorescu, F. Blaabjerg, and P. Rodríguez. "Grid Voltage Synchronization for Distributed
Generation Systems Under Grid Fault Conditions", IEEE Transactions on
Industry Applications. Vol. 51, No. 4, pp: 3414:3425, Jul 2015. [12] D.C. Mazur, R.A. Entzminger, and J.A. Kay. "Enhancing Traditional
Process SCADA and Historians for Industrial and Commercial Power
Systems with Energy (Via IEC 61850)", IEEE Transactions on Industry Applications. Vol. 52, No. 1, pp: 76:82, Jan 2016.
[13] R. Yousefian, and S. Kamalasadan. "Design and Real-Time
Implementation of Optimal Power System Wide-Area System-Centric Controller Based on Temporal Difference Learning", IEEE Transactions on
Industry Applications. vol. 52, no. 1, pp: 395:406, Jan 2016.
[14] V. Kekatos, G. B. Giannakis, and R. Baldick, “Grid topology identification using electricity prices,” in Proc. IEEE Power & Energy
Society General Meeting, National Harbor, MD, Jul. 2014.
[15] Y.C. Chen, T. Banerjee, A.D. Domínguez-García, and V.V. Veeravalli. "Quickest Line Outage Detection and Identification". IEEE Trans on Power
Syst, vol. 31, no. 1, pp 749-758, Jan 2016.
[16] Wu, J. Xiong, J. Shi, Y. “Ambiguity group based location recognition for multiple power line outages in smart grids”. In Proceedings of the 2014
IEEE PES on ISGT, Washington, DC, USA. [17] S. Azizi and M. Sanaye-Pasand, “A straightforward method for wide-
area fault location on transmission networks,” IEEE Trans. Power Del., vol.
30, no. 1, pp 264–272, Feb. 2015. [18] H. Zhu and G. B. Giannakis, “Sparse Overcomplete Representations for
Efficient Identification of Power Line Outages,” IEEE Tran. on Power Syst,
vol. 23, no. 4, pp. 1644–1652, Oct. 2012. [19] M. Babakmehr, M.G. Simões, A. Al Durra, F. Harirchi and Qi. Han.
“Application of Compressive Sensing for Distributed and Structured Power
Line Outage Detection in Smart Grids”, in Proceeding with 2015 American Control Conference-ACC 2015, Chicago, Illinois, USA, July 2015.
[20] A.Arabali, M.Majidi, M.S.Fadali, and M.Etezadi-Amoli, “Line Outage
Identification-Based State Estimation in a Power System with Multiple Line Outages,” Electric Power Systems Research, Vol.133, pp.79–86, April 2016.
[21] Chen, J.C.; Li, W.T.; Wen, C.K.; Teng, J.H.; Ting, P. “Efficient
identification method for power line outages in the smart power grid”. IEEE Trans. Power Syst. 2014, vol: 29, pp: 1788–1800.
[22] H. Sedghi, E Jonckheere. Information and Control in Networks. In
Giacomo Como, Bo Bernhardsson, Anders Rantzer, editors, Lecture Notes in
Control and Information Sciences, vol:450, 2014, Springer Cham Heidelberg
New York Dordrecht London ISBN: 978-3-319-02149-2, Ch. 9, pp 277-297. [23] M.Babakmehr, M.G.Simoes, M.B.Wakin, and F.Harirchi. “Compressive
Sensing-Based Topology Identification for Smart Grids”.
DOI:10.1109/TII.2016.2520396, IEEE Trans on Industrial Informatics. [24] M.Majidi, M.Etezadi-Amoli, and M.S.Fadali, “A Novel Method for
Single and Simultaneous Faults Location in Distribution Networks,” IEEE
Transactions on Power System. Vol. 30, Issue 6, pp. 3368-3376, Nov. 2015. [25] M.Majidi, A.Arabali, and M.Etezadi-Amoli, “Fault Location in
Distribution Networks by Compressive Sensing,” IEEE Transactions on
Power Delivery. Vol. 30, Issue 4, pp. 1761-1769, Aug. 2015. [26] E. J. Candes, M. B. Wakin, and S. P. Boyd, Enhancing sparsity by
reweighted 𝑙1-minimization, J Fourier Anal Appl (2008) 14: 877–905. [27] R. D. Zimmerman, C. E. Murillo-Sanchez, and R. J. Thomas,
“Matpower: Steady-state operations, planning and analysis tools for power systems research and education,” IEEE Trans. on Power Systems, vol.26, no.
1, pp. 12–19, Feb. 2011.
[28] A. Abur and A.G Exposito. Power System State Estimation, Theory and Implementation. Marcel Dekker, 2004.
[29] A. J. Wood and B. F. Wollenberg, Power Generation, Operation, and
Control, 2nd ed. New York: Wiley, 1996. [30] A. Schellenberg, W. Rosehart, and J. Aguado, “Cumulant-based
probabilistic optimal power flow (P-OPF) with Gaussian and Gamma
distributions”, IEEE Trans. on Power Syst, vol. 20, pp. 773–781, May 2005. [31] E.J. Candes and M.B. Wakin, “An introduction to compressive
sampling”, IEEE Sig Proc Magazine, vol. 24, pp. 21–30, March 2008.
[32] M. B. Wakin. Compressive Sensing Fundamentals. In M. Amin, editor, Compressive Sensing for Urban Radar, pp. 1-47. CRC Press, Boca Raton,
Florida, 2014.
[33] M. Grant and S. Boyd. CVX: Matlab software for disciplined convex programming, version 1.1. September 2007.
[34] S. Becker, J. Bobin, and E. J. Candes, NESTA: A fast and accurate first-
order method for sparse recovery, SIAM J. Imaging Sci., 4 (2011), pp. 1–39. [35] Yang, C., Guan, Z. H., Liu, Z. W., Chen, J., Chi, M., & Zheng, G. L.
“Wide-area multiple line-outages detection in power complex networks”.
International Journal of Electrical Power & Energy Systems, 79, 132-141, 2016.
[36] M. Babakmehr, M. G. Simoes, M. B. Wakin, A. Durra, and F. Harirchi,
827 “Smart grid topology identification using sparse recovery,” in Proc.
IEEE IAS Annu. Meeting, Oct. 2015, pp. 1–8.
Mohammad Babakmehr (S’14) received the
B.S. degree in electrical engineering in 2008 from Central Tehran University and the M.Sc. degree in
Biomedical-Bioelectric engineering in 2011 from
the Amirkabir University of Technology, Tehran, Iran. He is currently a Ph.D. degree candidate in
the Department of Electrical Engineering and
Computer Science, Colorado School of Mines (CSM), Golden. He has been with the Center for
the Advanced Control of Energy and Power
Systems since 2013. His research interests include smart grid technologies, compressive sensing, advance signal processing and control theory.
Marcelo G. Simões (S’89–M’95–SM’98-F’15)
Marcelo Godoy Simões received a B.Sc. degree from the University of São Paulo, Brazil, an M.Sc.
degree from the University of São Paulo, Brazil,
and a Ph.D. degree from The University of Tennessee, USA in 1985, 1990 and 1995
respectively. He received his D.Sc. degree (Livre-
Docência) from the University of São Paulo in 1998. Dr. Simões was an US Fulbright Fellow for
AY 2014-15, working for Aalborg University,
Institute of Energy Technology (Denmark). He is currently with Colorado School of Mines. He has been elevated to the grade
of IEEE Fellow, Class of 2016, with the citation: "for applications of artificial intelligence in control of power electronics systems."
0093-9994 (c) 2016 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.
This article has been accepted for publication in a future issue of this journal, but has not been fully edited. Content may change prior to final publication. Citation information: DOI 10.1109/TIA.2016.2574767, IEEETransactions on Industry Applications
Miachael B. Wakin (S’01–M’06–SM’13) Michael B. Wakin is the Ben L. Fryrear Associate Professor
in the Department of Electrical Engineering and
Computer Science at the Colorado School of Mines (CSM). Dr. Wakin received a Ph.D. in Electrical
Engineering from Rice University. He was an NSF
Mathematical Sciences Postdoctoral Research Fellow at Caltech from 2006-2007 and an Assistant
Professor at the University of Michigan from 2007-
2008. His research interests include sparse, geometric, and manifold-based models for signal
processing and compressive sensing. Dr. Wakin received the NSF CAREER
Award and also received the CSM Excellence in Research Award for his research as a junior faculty member.
Ahmed Al-Durra (S'07-M'10-SM'14) received the
B.S., M.S., and PhD in Electrical and Computer
Engineering from the Ohio State University in 2005, 2007, and 2010, respectively. He is an
Associate Professor in the Electrical Engineering
Department at the Petroleum Institute, Abu Dhabi, UAE. His research interests are application of
estimation and control theory in power system
stability, Micro and Smart Grids, renewable energy, and process control. He has published over
70 scientific articles in Journals and International Conferences. Dr. Ahmed has successfully
accomplished several research projects at international and national levels.
He is the co-founder of Renewable Energy Laboratory at the Petroleum Institute.
Farnaz Harirchi (S’14) received the B.S. degree in
electrical engineering in 2008 from Central Tehran
University and the M.Sc. degree in Electrial and
Electronic engineering in 2011 from the Iran
University of Science and Technology, Tehran,
Iran. She is currently a Ph.D. degree candidate in the Department of Electrical Engineering and
Computer Science, Colorado School of Mines
(CSM), Golden. She has been with the Center for the Advanced Control of Energy and Power
Systems since 2012. Her research interests include renewable energy
aggregation, smart Microgrids, power electronics and, intelligent control for high-power electronics applications.