300

IEEE TRANSACTIONS ON SMART GRID, VOL. 7, NO. 1, JANUARY 2016

Power System Real-Time Monitoring by Using

PMU-Based Robust State Estimation Method
Junbo Zhao, Student Member, IEEE, Gexiang Zhang, Member, IEEE, Kaushik Das, Student Member, IEEE,
George N. Korres, Senior Member, IEEE, Nikolaos M. Manousakis, Member, IEEE,
Avinash K. Sinha, Member, IEEE, and Zhengyou He, Senior Member, IEEE

AbstractAccurate real-time states provided by the state estimator are critical for power system reliable operation and
control. This paper proposes a novel phasor measurement
unit (PMU)-based robust state estimation method (PRSEM) to
real-time monitor a power system under different operation conditions. To be specific, an adaptive weight assignment function to
dynamically adjust the measurement weight based on the distance
of big unwanted disturbances from the PMU measurements is
proposed to increase algorithm robustness. Furthermore, a statistical test-based interpolation matrix H updating judgment strategy is proposed. The processed and resynced PMU information
are used as priori information and incorporated to the modified
weighted least square estimation to address the measurements
imperfect synchronization between supervisory control and data
acquisition and PMU measurements. Finally, the innovation
analysis-based bad data (BD) detection method, which can handle
the smearing effect and critical measurement errors, is presented.
We evaluate PRSEM by using IEEE benchmark test systems and
a realistic utility system. The numerical results indicate that, in
short computation time, PRSEM can effectively track the system
real-time states with good robustness and can address several
kinds of BD.
Index TermsInterpolation matrix, phasor measurement
unit (PMU), prior knowledge, state estimation (SE), weighted
least squares (WLSs).

I. I NTRODUCTION
INCE THE pioneering work by Schweppe and Wildes [1],
power system state estimation (SE) has been much more
important in todays energy management systems (EMSs). It

Manuscript received August 25, 2014; revised January 12, 2015 and
March 24, 2015; accepted May 6, 2015. Date of publication May 25, 2015;
date of current version December 19, 2015. This work was supported in
part by the National Natural Science Foundation of China under Grant
61170016 and Grant 61373047, in part by the Program for New Century
Excellent Talents in University under Grant NCET-11-0715, and in part
by Southwest Jiaotong University (SWJTU) under Project SWJTU12CX008.
Paper no. TSG-00843-2014.
J. Zhao, G. Zhang, and Z. He are with the School of Electrical
Engineering, Southwest Jiaotong University, Chengdu 610031, China (e-mail:
junbo_zhao@hotmail.com; zhgxdylan@126.com; hezy@swjtu.edu.cn).
K. Das is with Wind Energy Systems, Department of Wind Energy,
Technical University of Denmark, Roskilde 4000, Denmark (e-mail:
kaushik_msit@hotmail.com).
G. N. Korres and N. M. Manousakis are with the School of Electrical
and Computer Engineering, National Technical University of Athens, Athens
15780, Greece (e-mail: gkorres@cs.ntua.gr; manousakis_n@yahoo.gr).
A. K. Sinha is with the Department of Electrical Engineering, Indian
Institute of Technology (IIT) Kharagpur, Kharagpur 721302, India (e-mail:
aksinha@ee.iitkgp.ernet.in).
Color versions of one or more of the figures in this paper are available
online at http://ieeexplore.ieee.org.
Digital Object Identifier 10.1109/TSG.2015.2431693

provides accurate and continuously updated snapshots of the

system real-time states, which enables EMS to perform various important control and planning tasks such as optimizing
power flows, voltage stability analysis and bad data (BD)
detection/analysis, etc. [2].
In view of the smarter grid development, more attention
should be paid on SE research to meet the challenges that
provide the smart grid functionalities. This implies the incorporation of novel concepts and methods, such as those related
to advanced measurement technologies like phasor measurement units (PMUs), which make possible the near real-time
monitoring of the power grid [3]. However, the bus voltage and
branch current phasor measurements provided by PMUs are
sometimes incompatible with conventional supervisory control
and data acquisition (SCADA) systems because of significantly different sampling rates between synchronized PMU
measurements and nonsynchronized conventional SCADA
measurements [4]. On the other hand, PMUs cannot be
installed at every bus of a large power system due to economic
constraints [5]. As a consequence, the incorporation of the limited number of PMUs to the conventional SCADA-based SE
has been a major research subject in recent years.
Several SE methods incorporating SCADA and PMU measurements have already been extensively studied and can be
divided into two categories [3], [6]: a single state estimator, where PMU measurements are mixed with the traditional
SCADA measurements [7][13]; a two-stage scheme, where
the state obtained from the traditional SCADA measurements
is improved by using a second estimator that employs PMU
measurements [14][21]. The main idea of the former method
is to relate the voltage phasors in rectangular form to the
conventional state vector x in polar form through a simple
nonlinear transformation. Then, a single estimator, static or
dynamic, that incorporates both conventional and PMU measurements can be derived based on an augmented measurement
model. As compared to the SCADA-based SE methods, the
performance of these methods can be enhanced by using highprecision PMU measurements. However, these methods could
not be taken into account the advantage of faster data rate from
PMUs as PMU measurements between two SCADA measurements are dropped, and hence, cannot track the system
dynamics. So, the two stage scheme is proposed to utilize the
fast measurements from PMUs and on the other hand to reduce
the SE computation time, especially for large-scale power
networks. Particularly, interesting are the works [15][21].

c 2015 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission.
1949-3053 
See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.

ZHAO et al.: POWER SYSTEM REAL-TIME MONITORING BY USING PRSEM

In [15], a hierarchical scheme for distributed SE using PMU

measurements is first introduced. References [16] and [17]
improved this approach by considering a large number of tie
lines among subsystems. In [18], a distributed state estimator which locally estimates the states at substation level is
proposed for systems completely observable through PMU.
Inclusion of PMU measurements in multilevel state estimators
has also been considered in [6]. The estimation procedure [19]
is composed of two stages. A PMU-based linear estimator is
implemented at the first stage, processing only available in
PMU measurements. A traditional nonlinear estimator is used
and fused by multisensor data fusion theory at the second
stage, processing the estimated states of the first stage and
the available SCADA measurements. A hybrid state estimator
was proposed in [20] where a classical state estimator is incorporated along with a linear state estimator. A more extensive
literature survey and related references may be found in [21].
Note that, the mentioned papers above focus on power system steady SE using both SCADA and PMU measurements
and cannot be used to track system dynamics. To address this
problem, we have proposed a real-time hybrid SE method
incorporating SCADA and PMU measurements in [22], as
an extension of [20]. The main contribution of that method
is to classify the available measurement set into four subsets, then giving different weighting factors to these four
measurement sets artificially, resulting an estimation accuracy improvement comparing with method [20]. But, there
are still some problems left, which can be summarized as
follows. First, the weight is artificially assigned according to
the distance between the measured bus and the fault, which
may not be reasonable and feasible, especially under multiple
fault condition. Second, states of PMU unobservable buses
are interpolated from the live PMU data using a precomputed
interpolation matrix between two SCADA scans and the interpolation matrix will be updated when both SCADA and PMU
measurements arrive together. When the fault occurs between
two scans of SCADA measurements, the states of the PMU
unobservable buses, which are interpolated from the PMU
data, will be deviated from their true values, resulting to inability tracking of the systems dynamic changes. Third, the BD
detection and identification is not analyzed and discussed.
In order to deal with the problems mentioned above and
further improve the algorithm robustness and BD detection
and identification performance of our previous algorithm [22],
this paper proposes a novel PMU-based robust state estimation method (PRSEM), where an adaptive weight assignment function to dynamically adjust the measurement weight
based on the distance of big unwanted disturbances from
the PMU measurements, is proposed to increase algorithm
robustness. Furthermore, a statistical test-based interpolation
matrix H updating judgment strategy is proposed. The processed and resynchronized PMU information are used as priori
information and incorporated to the modified weighted least
square (WLS) estimation to address the imperfect synchronization between SCADA and PMU measurements. Finally,
the innovation analysis-based BD detection method, which can
handle the smearing effect and critical measurement errors, is
presented.

301

The rest of this paper is organized as follows. Section II

presents the proposed PRSEM. In Section III, experiments are
presented and the results are analyzed. Finally, Section IV
concludes this paper.
II. PRSEM
A. PMU Configuration and Measurement Partition
In this paper, the power grid is partitioned, according to
the PMUs configuration, into two areas: PMU observable
area (O) and PMU unobservable, but SCADA observable
area (U). Since the voltage of the generation and load buses
may undergo large variations when a system undergoes a disturbance [23], [24], it is a critical factor in monitoring and the
control of power system. The PMU configuration is to place
a certain number of PMUs at generation or important load
buses, given a desired depth of unobservability. Method [20]
is used to check the system observability under the specific PMU deployment condition. To improve the estimation
accuracy during the disturbance, PRSEM partitions the measurement set into four subsets based on a disturbance location
detection method, presented in the next section, and an adaptive measurement weight assignment function is proposed to
dynamically adjust their weights. Let us consider the IEEE
14-bus test system for example. We assume that the system
is completely observable through SCADA measurements and
there is only one PMU installed at bus 2, dividing the system
into areas O (buses 15) and U (buses 614). Furthermore, a
fault occurs between two SCADA scans at bus 6, which is not
directly observable by PMU. In this case, the proposed disturbance detection method is used to judge the disturbance at
bus 6. Four measurement sets are defined: PMU measurements
(buses 15), SCADA measurements at the faulted bus (bus
6), SCADA measurements directly connected to the faulted
bus (buses 1113), and the remaining SCADA measurements
(buses 710 and 14). The proposed weight function is adopted
to assign least weight to SCADA measurements at faulted bus
6, lower weights to measurements at buses 1113, moderate
weights to measurements at buses 710 and 14, and highest
weights to PMU measurements.
B. PMU-Based Robust State Estimation Method
In this section, a novel PRSEM, which is an extended
version of our previous work [22], is proposed. The main
ideas of PRSEM are the adaptive weight assignment function to dynamically adjust the measurement weight matrix of
area U when the system encounters a big unwanted disturbance and the modified WLS method with prior knowledge
calculated at previous PMU sample to update the interpolation matrix instead of updating it when a new set of SCADA
measurements becomes available [20], [22].
1) Mathematical Model: From the bus frame of reference,
the nodal equations in a power system can be expressed in the
following matrix form:


  
YOO YOU VO
IO
=
(1)
IU
YUO YUU VU
where IO and VO , IU and VU are the bus injection and voltage
vector of areas O and U, respectively; YOO and YUU are the

302

IEEE TRANSACTIONS ON SMART GRID, VOL. 7, NO. 1, JANUARY 2016

self-admittance of observable and unobservable buses, respectively; YOU and YUO are the mutual admittances of the lines
connecting observable and unobservable buses. It is assumed
that the current injection IU at the unobserved buses can
be modeled as equivalent load admittances at this particular
operating point [23], [25]. Then, in a system with NU unobserved buses, the equivalent load admittance vector YU at the
unobserved buses can be expressed as


  
[YU ] = IU /VU = Si |VU (i)|2
(2)
where Si is the complex power of node i and Si = Pi + jQi ;
VU (i) is the voltage magnitude of node i(i = 1, 2, . . . , NU ).
Substituting (2) into (1), we can get


  
YOO YOU VO
IO
=
(3)
0
YUO
YT
VU
where YT = YUU + YL ; YL = diag[YU ] is a NU NU diagonal
matrix.
Performing the indicated operation on the lower row of
matrix (3) yields
YUO VO + YT VU = 0.

(4)

The relationship between the vector of PMU unobservable

buses and the vector of PMU observable buses can be
expressed from the above equation as follows:
VU = YT 1 YUO VO .

(5)
1

If NO is the number of observable buses, the product YT

YUO expresses the NU NO sparse matrix of the admittance
elements. Let H = YT 1 YUO the matrix of interpolation
coefficients. The model for the interpolation of unobserved
buses in matrix form can be expressed as
VU = H VO .

(6)

Note that, if a line outage inside the area U or significant

load change happens between the arrival of the next a set of
PMU measurements, the above stated matrix of interpolation
coefficients needs to be updated.
2) Interpolation Coefficient Matrix Calculation: Any set
of interpolation coefficients H is calculated using (6) and
assuming a reference operating point, say 0. That is

1

YUO .
(7)
H 0 = YT 0
If the system changes its operating point, the new interpolation
coefficient matrix will be
H = H 0 + H
1

(8)

where H = YT YT H, which can be calculated by

taking the differential of both sides of H = YT 1 YUO , is an
error matrix caused by the line outage inside the area U or a
/|V |2 (S0 ) /|V 0 |2 ],
significant load change; YT = diag[SU
i
U
U

where SU and Vi are the conjugate complex power injection and voltage of unobservable buses at the present
0 and V 0 are the
configuration, respectively, whereas SU
U
corresponding initial operating configuration, respectively.
0 and V 0 are known, the important step is to
Since SU
U
and V .
estimate SU
i

3) Proposed Interpolation Coefficient Matrix Updating

Method: If there is no huge change in system topology or
load between two SCADA scans, the states of the area U can
be easily calculated by (6) and the interpolation coefficient
matrix will only be updated when the new SCADA and PMU
measurements arrive. However, if a line outage inside the area
U or a significant load change occurs between two SCADA
scans, the states of the PMU unobservable buses, which are
interpolated from the live PMU data, will deviate from the
true system states in area U resulting in being unable to track
the system dynamic changes.
In [20], H is updated only when SCADA and PMU measurements arrive simultaneously ignoring the occurrence of
disturbance between two SCADA scans, which is not able to
track the real-time states. To solve this problem, in our previous work [22], H is computed at every PMU sample between
two SCADA scans and also updated when both SCADA and
PMU measurements come together resulting in tracking ability improvement. However, updating H at every PMU sample
is not reasonable and will increase the computation burden.
To address this problem and increase the algorithm robustness when the system encounters big disturbance, a modified
WLS method combined with the H updating judgment strategy
for estimation of the area U states, considering the recursive priori state information is proposed. The related time
sequence of this method is presented in Fig. 2. The main idea
of this method is first to divide the two SCADA measurements sampling time interval into M PMU samples according
to the sample rate of PMU. Note that, these PMU sampling
information from PMU observable area will be processed and
resynchronized by the method in [26] before the execution
of SE and state interpolation of PMU unobservable area. The
priori state information of area U is calculated at the previous PMU sample by using the interpolation coefficient matrix,
which is further used to check the disturbance through our
proposed judgment criteria; if a line outage inside the area
U or a significant load change occurs, this priori state information is incorporated into the WLS-based SE algorithm to
and V of area U, followed by H updating; othestimate SU
i
erwise, interpolate the state of area O at every PMU sample
without H updating. Note that, whether a line outage inside
the area U or a significant load change happens or not, H will
be recomputed based on topology and state information when
the new SCADA and PMU measurements arrive.
The measurement model of the power system can be
expressed as
zk = h(xk ) + vk

(9)

where zk is the observed measurements vector consisted of

node-voltage magnitudes, active or reactive line flows and
nodal injections [1]; xk represents the state vector (nodevoltage magnitudes and phase angles) at each time sample k;
h() represents the vector of nonlinear measurement functions;
vk is the vector of Gaussian measurement error with zero mean
and covariance matrix Rk .
In practice, the PMU and SCADA measurements cannot
be directly combined into the hybrid SE due to the measurements imperfect synchronization [26]. As indicated in

ZHAO et al.: POWER SYSTEM REAL-TIME MONITORING BY USING PRSEM

Schweppe and Wildes pioneering work [1], the SCADA measurements are assumed to be taken simultaneously providing
a true snapshot of the system state, therefore, the SE is
performed to get the states of this snapshot. In this paper,
we also assume that the SCADA measurements are synchronized as the most of work in literature. The method [26] can
be used to process and resynchronize the imperfect synchronized PMU information from the PMU observable area. Thus,
the priori information can be calculated through matrix H
using these processed and resynchronized PMU information.
Note that, this priori information is important for improving measurement redundancy of PMU unobservable area. In
our former conference paper [22], this case is not considered. However, it should be taken into account for practical
power systems since some SCADA measurements may not be
accessible due to communication loss, leading measurement
redundancy greatly reduced. Finally, the priori information is
incorporated to the modified WLS estimation, whose objective
function is
min J(xk ) =
xk

1
[zk h(xk )]T Wk [zk h(xk )]
2
1
+ (xk 
xk )T P1
xk ) (10)
k (xk 
2

2 , 2 . . . 2 ])1 is a diagwhere Wk = Rk 1 = (diag[k,1

k,2
k,N
onal matrix of weighting factors; 
xk = x(kp) = Hkp
Ekp,O = (H 0 + Hkp ) Ekp,O denotes the vector
of the priori state values; Pk is the corresponding NU NU
covariance matrix; k k + ip, i = 0, 1, . . . , M.
The above objective function is solved through the
GaussNewton method [1], [2], leading to the following
iterative solution scheme:


x = QT Wk z + P1
QT Wk Q + P1
(11)
k
k x

where Q is the Jacobian matrix of h(x) at a given point xk ;

xk ; Pk = E[(xk 
xk )(xk 
xk )T ].
z = z h(xk ); x = xk 
The solution of (11) yields the vector of increments to the
states x, so that the updated state vector is obtained as
xk+1 = xk + x.

(12)

The convergence of the iterative procedure is attained when

x becomes smaller than a prespecified tolerance.
a) H updating judgment: It is not possible and reasonable to solve the WLS and update H at each PMU sample
since the time needed to solve WLS may be larger than PMU
sample rate for some systems. Note that, either a line outage inside the area U or a significant load change occurs,
the power flow measurements nearby the disturbance bus will
have significant variations compared with its steady operation value. Inspired by that, this paper proposes a power flow
variation-based metric to judge whether the matrix H needs to
be updated or not.
xk the priori state calculated at the previous PMU
Let 
xk and 
sample (if H is just updated, this state will be the previous SE
result) and the present state calculated at present PMU sample
using (6) in area U, respectively. The error covariance matrix

303

1
of 
xk can be derived as Sk = (QTk Wk Qk + P1
k ) , while, error
covariance matrix of 
xk is

cov(
xk ) = cov(Hk Vk ) = Hk cov(Vk ) HkT = Hk Lk HkT
(13)
where Lk is the error covariance matrix of the PMU phasors
at time sample k; Qk is the Jacobian matrix. Then, the priori and present estimated power flow can be calculated as
xk ) and 
xk ), respectively. The corresponding

zk = h(
zk = h(
error covariance matrices are k = Qk Sk QTk and
k = Qk Hk Lk HkT QTk = (Qk Hk ) Lk (Qk Hk )T . (14)
Proposition 1: Ideally, the innovation measurements residual (i) = 
z(i) 
z(i) at time sample k should be normally
distributed with zero mean and covariance K = + .
Proof: In this paper, as well as in many literature concerning the incorporation of PMU measurements into SE, SCADA,
and PMU measuring systems are regarded as distinct classes
of sensors, obtained via separate metering channels and used
to get independent estimation results. Therefore, measurement
errors presented in each of these measurement sets will be
considered uncorrelated. Besides, the measurement errors of
PMUs at different time sample are assumed to be uncorrelated.
The calculated 
z(i) using SCADA measurements and previous
time sample PMU measurements is independent from the estimated 
z(i) using the present PMU measurements. Thus, the
covariance of the residual is
cov(
z(i) 
z(i)) = cov(
z(i)) + cov(
z(i)) = + . (15)
On the other hand, the gross error of SCADA and PMU
measurements are usually assumed to be normally distributed
with zero mean. So, we can easily derive: [
z(i) 
z(i)] = 0,
where  is the expectation operator.
Finally, the residual vector is normalized and submitted
to the following statistical test to detect the existence of
disturbance:
|r(i)|
threshold
(16)
N (i) =
K (i)

where K (i) = K(i, i) is the standard deviation of the ith

component of vector r. The threshold is determined by the
test confidence, i.e., the detection threshold for the test can be
set to 3 with a 99.7% confidence level.
As we know, under normal system operation conditions,
the power flow changes a little due to the continuous small
variation of loads and the detection result will be negative.
However, when a line outage inside the area U or a significant
load change happens, the maximal normalized residual will
have large probability to be larger than the defined threshold,
i.e., 3 with 99.7% confidence level.
b) Robustness enhancement: Under normal conditions,
a power system changes slowly (H 0) with time due to the
continuous small variation of loads and we can obtain accurate system real-time states of area U through the interpolation
coefficient matrix H. However, if the power system encounters a big unwanted disturbance such as large load change or
topology change, the performance of the SE algorithm will

304

IEEE TRANSACTIONS ON SMART GRID, VOL. 7, NO. 1, JANUARY 2016

Fig. 2. State sequence of the modified WLS-based estimation method in the

two SCADA measurements sample time.

Fig. 1.

Partitioning for the IEEE 14-bus system.

be significantly affected deteriorating the computation performance, or even worse, causing a divergence of the algorithm.
Therefore, the updating of H cannot be accomplished in time
and the accurate states of area U cannot be obtained. In addition, when the disturbance happens, the measurement weight
should be dynamically assigned to reflect the measurement
accuracy. The measurements that are near to the location of
disturbance sources will be assigned with a lower weight,
whereas those measurements that are far away from the disturbance sources will approximately maintain their weights had
before the occurrence of disturbance. Generally, it is hard to
dynamically adjust the specific weight values according to the
measurement accuracy. In order to overcome these problems,
we propose an adaptive weight assignment function, that is

xk xk 2 .
Wknew =Wk exp 

(17)

Note that, the measurement set comprises voltages, active and

reactive power injections, and flows. Since the active and reactive power measurements have little impact on the voltage
and angle according to the PQ decomposition strategy, in
the new weight function, the new weights for active power
injections and flows are determined by 
k k 2 , whereas
2

Vk Vk is used to determine the weights of the reactive
power injections and flows in (17).
c) Illustrate examples for disturbance detection: Since
the disturbance in PMU observable area can be easily located
using the methods in the literature, only the disturbance in
area U is considered and discussed.
i) Single disturbance: In this case, when a big unwanted
disturbance happens, the strategy in H updating judgment
is first used to detect the existence of disturbance; then all
the power flow, which satisfy (16), will be chosen out; finally,
the common bus among these power flow will be marked as the
disturbances location. For example, in Fig. 1, the disturbance
existence is first verified, then the power flows P65 , P611 ,
P613 , and P612 beyond the defined metric are chosen out
and finally the common node 6 is determined as the location
of disturbance.

ii) Multidisturbances: In this case, the multidisturbances

can be actually classified as many single disturbances and
the method presented above can be used to determine the
disturbance source in each disturbance area. Consider a two
disturbance sources case for example. In the IEEE 14-bus test
system, two temporary three phase to ground faults happened
at buses 6 and 7 simultaneously. By using the single disturbance detection method, we can find that power flows P65 ,
P611 , P613 , P612 , P47 , P78 , and P79 have violated the
defined threshold indicating disturbances at buses 6 and 7,
respectively.
After that, the adaptive weight assignment function is
adopted to dynamically adjust the measurement weight according to the measurement accuracy. For example, if a big
unwanted disturbance happens, the values of 
xk will have significant deviation from the present states xk , increasing the
value 
xk xk 2 , and therefore, the weighting factor should be
reduced by using the negative exponential function, which suppress the influence of the disturbance and ensure that PRSEM
will not diverge. In other words, the measurements that are
near to the location of disturbance sources are dynamically
assigned a lower weight, while those are far away from the
disturbance sources will approximately maintain the weight
having before the occurrence of disturbance because the value
of 
xk xk 2 output approaches zero.
C. Bad Data Analysis
In this paper, when both SCADA and PMU measurements arrive simultaneously, i.e., at time sample k + 1, H is
updated and the interpolated state information using present
PMUs is used to detect the BD. The idea is similar to the
anomaly detection and processing based on the innovation
analysis in dynamic SE [27]. At time sample k + 1, the state
xk+1 is calculated by using the WLS method using present
SCADA measurements and priori state information, followed
by estimated measurement calculation zk+1 = h(xk+1 ), whose
error covariance matrix is Tk+1 = QSk+1 QT , where Sk+1 =
1 is the covariance of the SE error. At the
(QT Wk+1 Q + P1
k+1 )
same time, the present PMU measurements are used to get the
present state information 
xk+1 , followed by the present measurement calculation 
zk+1 = h(
xk+1 ), whose error covariance

ZHAO et al.: POWER SYSTEM REAL-TIME MONITORING BY USING PRSEM

305

Fig. 4. Voltage magnitude and phase angle for bus 13 of the IEEE 30-bus
system considering a three phase fault.

Fig. 3.

Flowchart for the PRSEM algorithm.

T
matrix is Mk+1 = QP1
k+1 Q . Then, the residual vector can be
calculated as

rk+1 (i) = zk+1 (i) 

zk+1 (i)

(18)

which is approximately a white Gaussian process with zero

mean and covariance matrix Nk+1 = Sk+1 + Mk+1 . Residual
is normalized and submitted to the statistical test to detect
the BD
rN (i) = |r(i)|/N (i) threshold
(19)

where N (i) = N(i, i) is the standard deviation of the ith

residual. Note that, the method [28] will be used to detect and
process the bad PMU measurements if they occur, then the
valid PMU measurements are applied to detect the existence
of bad SCADA measurements.
In this paper, a positive indication of test suggests the
inconsistence between the expected information of the system
(the priori information) and the present data. Such anomaly,
i.e., loss of generation, sudden load change, or unscheduled
outage network configuration, does not cause high residuals
due to the incorporation of new measurement information.
Thus, the system configuration switching is confirmed when
test is positive while rN test is negative; otherwise, when
both test and rN test are positive, the BD is detected and
identified.
III. S IMULATION R ESULTS
The proposed PRSEM algorithm is evaluated by using the
IEEE 30- and 118-bus systems and a realistic utility system B.
The flowchart of the proposed algorithm is shown in Fig. 3. It
is assumed that both IEEE test systems are completely observable through SCADA measurements whereas some buses are

unobservable through PMUs. For simulations purposes, the

SCADA measurements are generated using a conventional
fast decoupled power flow program, whereas PMU data were
generated using a time domain transient stability program.
To simulate real scenarios, 2% and 0.2% random errors are
incorporated in the SCADA and PMU data, respectively.
A redundancy of 1.6 is considered for SCADA measurements
ensuring the complete topological observability of the system.
For the IEEE 30-bus system, five PMUs are placed at buses
2, 6, 10, 15, and 25, whereas for the IEEE 118-bus system 16
PMUs are placed at buses 8, 15, 26, 30, 38, 45, 52, 57, 63,
68, 74, 81, 90, 99, 108, and 117, respectively. For simulation
purposes, it is considered that PMU measurements arrive at
an interval of 20 ms, whereas SCADA measurements arrive
at an interval of 4 s. All the tests are performed in MATLAB
environment using a 2.5 GHz, 8 GB of RAM, Intel Core i5
computer.

A. Fault Analysis
In order to demonstrate the effectiveness and robustness of
the proposed PRSEM algorithm, a temporary (0.1 s) three
phase to ground fault in IEEE 30-bus system is simulated at
the bus 13, which is unobservable through PMUs. Since the
dynamic system states can be easily captured in area O by
real-time PMU measurements, only the node dynamic states
in area U are shown here. To validate the proposed PRSEM
algorithm, its results are compared with those of estimation
methods in [20] and [22] in terms of the true value obtained
from time domain transient stability solution.
The result of the estimated states of the faulted bus 13 is
presented in Fig. 4. From this figure, it is clear that PRSEM is
much better than the other two methods in [20] and [22] and
it provides results almost identical to the actual states of the
system. It can also be seen that PRSEM can quickly track
the dynamic state change from the very beginning, whereas
the two other methods need more time to track the new
states, which means that the ability of PRSEM to dynamically track the system state change is superior compared with

306

IEEE TRANSACTIONS ON SMART GRID, VOL. 7, NO. 1, JANUARY 2016

TABLE I
R ESULTS FOR C ASE 1 IN IEEE 30- AND 118-B US T EST S YSTEMS

TABLE II
R ESULTS FOR C ASE 2 IN IEEE 30- AND 118-B US T EST S YSTEMS

Fig. 5. Voltage magnitude and phase angle for bus 30 of the IEEE 30-bus
system considering a three phase fault.

the other two methods. This is because of adopting the proposed H updating strategy to timely update H so that system
dynamic state change can be captured and the adaptive weight
assignment function to improve the algorithm robustness. For
example, when a new set of PMU measurements are received,
all the power flows through all the lines connected between
any observable and unobservable buses are calculated, then the
judgment metric in (16) is used to determine the existence of
disturbance. In this case, P1213 and Q1213 are larger than the
predefined threshold, indicating the existence of disturbance at
node 12. After that, the dynamic weight factors are selected
for different sets of measurements by using the proposed adaptive weighting function in the WLS procedure. For example,
the SCADA measurement corresponding to the faulted bus 13
is the least weighted (the value 
xk xk 2 is the largest). The
SCADA measurements for buses 4 and 12, which are directly
connected to the faulted bus, will be given low weight, whereas
the remaining SCADA measurements will be given moderate
weight. Finally, H is updated by using this modified WLS.
Fig. 5 shows the results of a remote bus 30 (not directly connected to faulted bus and not observable from PMUs). From
this figure it is obvious that the PRSEM algorithm can also
effectively track the system dynamic state change, compared
to the other two methods, which means that the over-fitting of
dynamic weights does not appear. So, the introduced adaptive
dynamically adjustment weighting function is reasonable and
effective.
B. Bad Data Analysis
To demonstrate the feasibility of PRSEM for BD detection under different BD condition, two cases are considered in
the IEEE 30- and 118-bus test systems, whose measurements
configuration can be found in [29]. The detection threshold
adopted for normalized residual tests is 3.0 (99.7% confidence
level).
1) Case 1: For the IEEE 30 system, multiple interacting
and conforming BD in real-power injection at bus 30
and real-power flow in line 3027 is considered; while
for the IEEE 118 system, multiple noninteracting BD in

TABLE III
D IFFERENT PMU C ONFIGURATIONS AT U TILITY S YSTEM B

reactive power flow in line 11876 and reactive power

flow at buses 103110 is considered.
2) Case 2: For the IEEE 30 system, single BD is introduced in reactive power injection at bus 24, a critical
measurement; while for the IEEE 118 system, a critical
measurement in the reactive power flow in line 6359
is considered.
Note that, the BD is set up by adding gross measurement
errors of 0.3 p.u. to line flows and bus injections. Obtained
results from PRSEM for cases 1 and 2 in the IEEE 30and 118-bus test systems are presented in Tables I and II,
where NBD represents the nonbad data. Besides, only three
largest normalized residuals are shown here due to the space
limitation.
It can be observed that the proposed BD detection method
effectively discriminates bad measurements from NBD for
both test systems. To be specific, it is apparently clear from
Table I for case 1 that PRSEM could not only correctly detect
the multiple interacting and conforming BD but also detect
multiple noninteracting BD by using the innovation-based normalized residual test. On the other hand, from Table II, we can
find that even when BD exist among the critical measurements,
which is difficult to be detected in the conventional BD detection method, the proposed method can still accurately detect
them. Once these BD are detected, we use the new SCADA
measurements calculated by the states interpolated by highprecision PMU measurements to replace them. A new SE run
is established to get the new system state.
C. Utility System B Test
The utility system B [20] is adopted to further test the effectiveness and robustness of PRSEM. This system includes 444
buses and 574 lines. In this test, the load change curve is

ZHAO et al.: POWER SYSTEM REAL-TIME MONITORING BY USING PRSEM

307

TABLE IV
I NDICES FOR F OUR M ETHODS U NDER D IFFERENT P HASED I NSTALLATION

Fig. 7.

Fig. 6.

Evolution of voltage and angle at bus 243 with time in case 2.

set the same as in [20]. Three different PMU configurations,

considering the communication constraint are summarized in
Table III. Note that, the installation of PMUs progressively
increases the observability but due to the communication constraints, it is not possible to totally observe the system. To
measure the performance of PRSEM as the system responds
to the load ramp, the following indices are used:
V =

Nu
 est

1
V V true 
i
i
Nu

Evolution of MVA line flow from buses 43 to 325 in case 2.

value of voltage, angle, and apparent flow, whereas Vitrue , itrue ,

and fitrue are the corresponding true values.
Table IV summarizes three performance indices for voltage
magnitude, angle, and flows under different PMU installations at utility system B, where CH represents the constant
H method in [20]. In general, the indices show improvements
in estimated state as the number of PMUs increases. We can
also find that PRSEM is the best method among the other
methods considering the estimation accuracy. Besides, these
indices show that the three methods, [22], [20], and PRSEM,
in which different least-squares estimation (LSE) update techniques are adopted will result in more accurate estimation of
voltages and flows on the unobserved buses and lines than the
constant H estimation method.
Figs. 6 and 7 show the evolution of voltage magnitude and
angle of unobserved bus 243 in response to the load ramp and
the evolution of megavolt ampere (MVA) line flow from buses
43 to 325 in case 2, respectively. It can be easily seen from
these two figures that PRSEM is better than the other methods.
On the other hand, the estimation error among PRSEM, [20],
and [22] is considerably small because of the LSE updating
technique, and also compared with the constant H estimation
method.

i=1

D. Assessment of Computational Effort

Nu
 est

1
 true 
=
i
i
Nu

f =

1
Nu

i=1
NL

 est

 f f true 
i

(20)

i=1

where V , , and f represent the mean of average voltage

magnitude, voltage angle, and line power flows errors, respectively; Nu and NL are the number of unobserved buses and
lines, respectively; Viest , iest , and fiest are the estimated mean

The computational performance of the proposed PRSEM

algorithm is compared with those of the regression weighted
least squares estimation (RWLS) method and the methods
in [20] and [22], where at every iteration, the last SE results
are used to initialize the iterative states in RWLS resulting
in computation reduction compared with the traditional WLS
method. Note that, in the simulation, SE is only performed
in area U to update H. Simulation results for different SE
methods are presented in Table V.

308

IEEE TRANSACTIONS ON SMART GRID, VOL. 7, NO. 1, JANUARY 2016

TABLE V
T IME R EQUIREMENT FOR F OUR SE M ETHODS

ACKNOWLEDGMENT
The authors would like to thank the Editor-in-Chief,
Professor J. Wang, the Editors, and the anonymous reviewers
for their insightful comments and suggestions, which greatly
helped in improving the quality of this paper.
R EFERENCES

From this table it is obvious that the computation time for

all of these methods are considerably small and can be compatible with real-time application under steady or dynamic
operation condition. Also, as the systems size increases, the
computation cost remains in reasonable values and is not
incrementally affected. On the other hand, PRSEM and [22]
take more computational time than the method [20] due to
the additional prior information computation or interpolation
coefficient matrix H updating. This time can be accepted
considering the great improvement to the ability of tracking the system dynamic states change. The time for PRSEM
is slightly smaller than that of method [22] comparing the
proposed H updating method with that in [22]. Furthermore,
as the number of PMUs is increased, the computation time
for all the methods decreases since the area U needed
for SE becomes smaller and the corresponding updated H
decreases.
It should be noted that with the increased size of system,
the communications should be a big concern if the centralized SE is adopted. However, if the distributed SE is used,
which is widely used for very large systems, the communication requirement is greatly reduced. This is because the system
is portioned into several small areas, the communication and
SE for these small areas will be no longer the main issue. On
the other hand, we can observe from Fig. 1 that the proposed
method is based on the system partitioning. This means the
distributed SE can be applied to address the measurements
communication problem.

IV. C ONCLUSION
This paper presents a novel PRSEM for real-time monitoring of power systems under different operation conditions. By
introducing a modified WLS-based interpolation matrix updating strategy, the real-time states of PMU unobservable area
can be calculated using limited high-precision PMU measurements. An adaptive weight assignment function to dynamically
adjust the measurement weight matrix based on the distance
of big unwanted disturbance from PMU measurements is
proposed to increase the algorithm robust. In addition, the
innovation by using the priori state information-based BD
detection method, which can address the smearing effect and
critical measurement errors, is also presented and analyzed.
Simulations on some IEEE benchmark test systems and a
realistic utility system verify the effectiveness and robustness of PRSEM even in the presence of a disturbance in the
unobservable part of the system and BD.

[1] F. C. Schweppe and J. Wildes, Power system static-state estimation,

part I, II, III, IEEE Trans. Power App. Syst., vol. PAS-89, no. 1,
pp. 120135, Jan. 1970.
[2] A. Monticelli, Electric power system state estimation, Proc. IEEE,
vol. 88, no. 2, pp. 262282, Feb. 2000.
[3] Y. F. Huang, S. F. Werner, J. Huang, N. Kashyap, and V. Gupta, State
estimation in electric power grids, IEEE Signal Process. Mag., vol. 29,
no. 5, pp. 3343, Sep. 2012.
[4] A. G. Phadke, J. S. Thorp, R. F. Nuqui, and M. Zhou, Recent developments in state estimation with phasor measurements, in Proc. IEEE/PES
Power Syst. Conf. Expo., Seattle, WA, USA, Mar. 2009, pp. 17.
[5] A. G. Phadke and J. S. Thorp, Synchronized Phasor Measurements and
Their Applications. New York, NY, USA: Springer, 2008.
[6] A. Gmez-Expsito, A. Abur, A. de la Villa Jan, and C. Gmez-Quiles,
A multilevel state estimation paradigm for smart grids, Proc. IEEE,
vol. 99, no. 6, pp. 952976, Jun. 2011.
[7] G. N. Korres and N. M. Manousakis, State estimation and bad data processing for systems including PMU and SCADA measurements, Elect.
Power Syst. Res., vol. 81, no. 7, pp. 15141524, Jul. 2011.
[8] S. Chakrabarti, E. Kyriakides, G. Ledwich, and A. Ghosh, Inclusion of
PMU current phasor measurements in a power system state estimator,
IET Gen. Transmiss. Distrib., vol. 4, no. 10, pp. 11041115, Oct. 2010.
[9] C. Bruno et al., Possibility of enhancing classical weighted least
squares state estimation with linear PMU measurements, in Proc. IEEE
Bucharest PowerTech, Bucharest, Romania, 2009, pp. 16.
[10] Z. H. Ga, A new state estimation model of utilizing PMU measurements, in Proc. Int. Conf. Power Syst. Technol., Chongqing, China,
2006, pp. 15.
[11] F. Chen, X. Han, Z. Pan, and L. Han, State estimation model and
algorithm including PMU, in Proc. 3rd Int. Conf. Elect. Util. Deregul.
Restruct. Power Technol., Apr. 2008, pp. 10971102.
[12] H. Xue et al., A dynamic state estimation method with PMU and
SCADA measurement for power systems, in Proc. Int. Power Eng.
Conf., Singapore, 2007, pp. 848853.
[13] A. Jain and N. R. Shivakumar, Impact of PMU in dynamic state estimation of power systems, in Proc. 40th North Amer. Power Symp.,
Calgary, AB, Canada, 2008, pp. 18.
[14] G. N. Korres and N. M. Manousakis, State estimation and observability
analysis for phasor measurement unit measured systems, IET Gener.
Transmiss. Distrib., vol. 6, no. 9, pp. 902913, Sep. 2012.
[15] L. Zhao and A. Abur, Multi-area state estimation using synchronized phasor measurements, IEEE Trans. Power Syst., vol. 20, no. 2,
pp. 611617, May 2005.
[16] M. Zhou, V. A. Centeno, J. S. Thorp, and A. G. Phadke, An alternative
for including phasor measurements in state estimators, IEEE Trans.
Power Syst., vol. 21, no. 4, pp. 19301937, Nov. 2006.
[17] W. Jiang, V. Vittal, and G. T. Heydt, A distributed state estimator utilizing synchronized phasor measurements, IEEE Trans. Power Syst.,
vol. 22, no. 2, pp. 563571, May 2007.
[18] E. Farantatos, G. K. Stefopoulos, G. J. Cokkinides, and
A. P. Meliopoulos, PMU based dynamic state estimation for
electric power systems, in Proc. IEEE Power Energy Soc. Gen.
Meeting, Calgary, AB, Canada, Jan. 2009, pp. 18.
[19] A. S. Costa, A. Albuquerque, and D. Bez, An estimation fusion
method for including phasor measurements into power system real time
modelling, IEEE Trans. Power Syst., vol. 28, no. 2, pp. 19101920,
May 2013.
[20] R. F. Nuqui, State estimation and voltage security monitoring using synchronized phasor measurements, Ph.D. dissertation, Dept. Elect. Eng.,
Virginia Polytech. Inst. State Univ., Blacksburg, VA, USA, Jul. 2001.
[21] A. Gmez-Expsito, A. de la Villa Jan, C. Gmez-Quiles, P. Rousseaux,
and T. Van Cutsem, A taxonomy of multi-area state estimation methods, Elect. Power Syst. Res., vol. 81, no. 4, pp. 10601069, Apr. 2011.
[22] K. Das, J. Hazra, D. P. Seetharam, R. K. Reddi, and A. K. Sinha, Realtime hybrid state estimation incorporating SCADA and PMU measurements, in Proc. IEEE PES Innov. Smart Grid Technol. Europe (ISGT),
Berlin, Germany, 2012, pp. 18.

ZHAO et al.: POWER SYSTEM REAL-TIME MONITORING BY USING PRSEM

[23] C. Gonzalez-Perez and B. F. Wollenberg, Analysis of massive measurement loss in large-scale power system state estimation, IEEE Trans.
Power Syst., vol. 16, no. 4, pp. 825832, Nov. 2001.
[24] M. Glavic and T. Van Cutsem, Reconstructing and tracking network
state from a limited number of synchrophasor measurement, IEEE
Trans. Power Syst., vol. 28, no. 2, pp. 19211930, May 2013.
[25] D. E. Barber, Applications of phasor measurements to real-time monitoring of a power system, M.S. thesis, Dept. Elect. Eng., Virginia
Polytech. Inst. State Univ., Blacksburg, VA, USA, Mar. 1994.
[26] Y. Peng, Z. Tan, A. Wiesel, and A. Nehorai, Power system state estimation using PMUs with imperfect synchronization, IEEE Trans. Power.
Syst., vol. 28, no. 4, pp. 41624173, Nov. 2013.
[27] F. Aminifar, M. Shahidehpour, M. Fotuhi-Firuzabad, and S. Kamalinia,
Power system dynamic state estimation with synchronized phasor measurements, IEEE Trans. Power Syst., vol. 63, no. 2, pp. 352363,
Feb. 2014.
[28] M. B. Do Couto Filho, J. C. S. De Souza, and M. A. R. Guimaraens,
Enhanced bad data processing through phasor-aided state estimator,
IEEE Trans. Power Syst., vol. 29, no. 5, pp. 22002209, Sep. 2014.
[29] A. Tarali and A. Abur, Bad data detection in two-stage state estimation
using phasor measurements, in Proc. IEEE PES Innov. Smart Grid
Technol. Europe (ISGT), Berlin, Germany, 2012, pp. 18.

Junbo Zhao (S13) is currently pursuing the

Ph.D. degrees with the Southwest Jiaotong
University, Chengdu, China, and the Bradley
Department of Electrical Computer Engineering,
Virginia Polytechnic Institute and State University,
Blacksburg, VA, USA.
His current research interests include theoretical
and algorithmic studies in power system state
estimation, power system operation and control,
signal processing, optimization method, and their
applications in robot control and smart grid.

Gexiang Zhang (M03) received the B.Sc. and

M.Sc. degrees in signal and information processing
from the Southwest University of Science and
Technology, Mianyang, China, in 1997 and 2002,
respectively, and the Ph.D. degree from Southwest
Jiaotong University, Chengdu, China, in 2005.
Since 2005, he has been a Professor with
the School of Electrical Engineering, Southwest
Jiaotong University, where he leads the Research
Group of Nature-Inspired Computation and
Smart Grid. His current research interests include
natural computing, smart grid, and robotics. He has published over 100
scientific papers in international journals or conferences. He was selected as
New Century Excellent Talents in University from the Chinese Ministry of
Education, Beijing, China.

Kaushik Das (S13) received the M.Tech. degree in

power system engineering from the Indian Institute
of Technology, Kharagpur, India, in 2011. He is currently pursuing the Ph.D. degree with the Technical
University of Denmark, Roskilde, Denmark.
He was with IBM, India Research Laboratory,
Bengaluru, India. His current research interests
include power system defence plans, integration
of renewable energy resources in grid, wide area
monitoring, and control of power systems using
synchrophasor measurements.

309

George N. Korres (SM05) received the Diploma

and Ph.D. degrees in electrical and computer engineering from the National Technical University of
Athens (NTUA), Athens, Greece, in 1984 and 1988,
respectively.
He is currently a Professor with the School of
Electrical and Computer Engineering, NTUA. His
current research interests include power system state
estimation, power system protection, and industrial
automation.
Prof. Korres is a Member of the International
Council on Large Electric Systems.

Nikolaos M. Manousakis (M14) received the

B.S. degree in electrical engineering from the
Technological Educational Institute of Piraeus,
Egaleo, Greece, in 1998, and the Dipl.Eng. and
Ph.D. degrees in electrical and computer engineering from the National Technical University of
Athens (NTUA), Athens, Greece, in 2003 and 2013,
respectively.
He is currently a Postdoctoral Researcher with
the School of Electrical and Computer Engineering,
NTUA. His current research interests include power
system state estimation and identification techniques, and phasor measurement
unit technology.

Avinash K. Sinha (M91) received the B.Sc. degree

in electrical engineering from Patna University,
Patna, India, in 1976; the M.Tech. degree from
the Indian Institute of Technology (IIT) Kharagpur,
Kharagpur, India, in 1978; and the Ph.D. degree
from the Birla Institute of Science and Technology,
Pilani, India, in 1983.
He joined IIT Kharagpur, in 1984, where he is currently a Professor with the Department of Electrical
Engineering. He was the Head of Department from
2007 to 2010. He is leading the Power Systems
Group and his team has developed a digital simulator for power systems.
He was a Visiting Professor with Washington State University, Pullman, WA,
USA, from 2001 to 2002 and in 2011. His current research interests include
power system analysis, simulation of power system dynamics, artificial intelligence applications to power systems, and synchro-phasor applications to
power systems.
Prof. Sinha was a recipient of the IBM Open Collaborative Faculty Award
for 2010 and 2011.

Zhengyou He (M09SM13) was born in China,

in 1970. He received the B.Sc. and M.Sc. degrees
in electrical engineering from Chongqing University,
Chongqing, China, in 1992 and 1995, respectively,
and the Ph.D. degree from Southwest Jiaotong
University, Chengdu, China, in 2001.
Since 2002, he has been a Professor with the
Department of Electrical Engineering, Southwest
Jiaotong University. His current research interests
include areas of signal process and information theory and its application in electrical power system,
and the application of wavelet transforms in power system.