A Two-Stage Orthogonal Estimator to Incorporate Phasor Measurements into

Power System Real-Time Modeling

Antonio Simoes Costa

Andre Albuquerque

Federal University of Santa Catarina

Florianopolis, Brasil
simoes@labspot.ufsc.br

Federal University of Santa Catarina

Florianopolis, Brasil
andrealbuquerque@ieee.org

Abstract - This paper proposes the use of an orthogonal

state estimator with capability to process a priori information
as a means to incorporate the benefits of synchronized phasor measurements into power system state estimation. The
proposed strategy promotes the enhancement of results provided by a conventional, SCADA-based estimator by means of
a post-processing stage, where those estimates are treated as a
priori information to be improved by the processing of PMU
data through the orthogonal algorithm. The procedure does
not interfere with the structure of conventional estimators,
which can be based on any available method. Computational
efficiency is ensured by the properties of the particular version of the orthogonal Givens rotations employed in the postprocessing stage, which easily accommodates the processing of
a priori information without extra computational cost. The application of the proposed methodology is illustrated through
several case studies conducted on the IEEE 14-bus, 57-bus and
118-bus test systems.
Keywords - Power System State Estimation, synchronized
phasor measurements, phasor measurement units (PMUs), orthogonal Givens rotations.

1 Introduction
During the last several decades, Power System State Estimation (PSSE) has established itself as the basic tool for
real-time modeling of large electric power networks. As the
emerging Smart Grid concepts expand previous paradigms
for power system operation and control, PSSE must evolve
to keep pace with the current trends [1]. This requires the
incorporation of new technologies to fulfill more stringent
accuracy and observability requirements posed to state estimators.
Conventional state estimators process data gathered by
SCADA systems by scanning remote terminal units (RTUs)
located at the power system substations. The advent of
the phasor measurement technology has made it possible
to accurately measure bus voltage and branch current phasors through Phasor Measurement Units (PMUs), something previously infeasible with SCADA. Therefore, the use
of phasor measurements in power system state estimation
has deserved much attention in recent years.
Despite the clear advantages of the phasor measurement
technology, it is unlikely that SCADA systems for power
system real-time modeling are to be simply replaced by
PMU installations in the near future. A number of arguments can be raised to support that claim, such as the still insufficient number of PMU measurements usually available

17th Power Systems Computation Conference

to provide full system observability; the fact that other applications which can take better advantage of the high PMU
sampling rate tend to hold preference in the deployment of
that equipment, and the existence of a complex and expensive SCADA infrastructure that should not be simply discarded.
A sensible solution is to combine the widespread availability of conventional measurements with the enhanced
quality of observations provided by the PMU technology,
available at certain points of the power network. For that
purpose, however, one faces the problem of how to conceive state estimation strategies able to benefit most from
both technologies. It is of course possible to devise state
estimators able to simultaneously process both SCADA and
PMU measurements [2, 3, 4]. However, as pointed out in
[5], this would require significant changes to existing State
Estimation software, since most estimators presently operating worldwide are based solely on SCADA measurements
and thus not prepared to process phasor data.
This paper proposes a new approach to take full advantage of the superior quality of PMU measurements without
the need of introducing changes to the structure of existing
estimators. As in [5, 6], this is accomplished through a postprocessing stage to the conventional estimator. Therefore,
the whole estimation procedure is composed of two modules: (a) a conventional, SCADA-based state estimator, and
(b) a PMU-based estimator, intended to enhance the quality
of state estimation and the degree of network observability
by processing only the available PMU measurements.
The proposed approach relies on the concept of a priori
state information, and for that reason it will be referred to
as APSI state estimation. It conceptually differs from other
methods previously proposed in the literature in two aspects:
i) SCADA-based state estimates obtained from the first
stage are treated by the PMU-based module as a
priori information, that is, initial knowledge about the
state variables, whose quality is expected to be improved by processing the available phasor measurements, and
ii) The estimation algorithm in the post-processing module is based on a particular version of orthogonal
Givens rotations which, in addition to enhanced numerical robustness, easily accommodates a priori information at virtually no extra computational cost
[7, 8].

Stockholm Sweden - August 22-26, 2011

This paper is organized as follows. Section 2 reviews the

background of power system state estimation and extends
the conventional normal equation approach in order to take
into account the availability of a priori state information.
Section 3 describes the three-multiplier version of Givens
rotations and shows how this variant of the orthogonal rotations makes provision to take a priori information into
account. The proposed method to enhance SCADA-based
estimates through the processing of phasor measurements
is discussed in Section 4, followed by some additional remarks related to the methods performance and possible improvements, in Section 5. Section 6 presents the results of
several case studies performed with three IEEE test systems.
Finally, concluding remarks are listed and discussed in Section 7.
2

State Estimation Background

2.1 Measurement Model and Weighted Least Squares Estimator

Consider a N bus power network for which m measurements are taken. Assuming a nonlinear model for the electrical network, the relationships between measured quantities and the n = 2N 1 state variables can be expressed
as:
z = h(x) +
(1)
where z is the m1 measurement vector, x is the n1 vector of state variables to be estimated, h(x) is the m 1 vector of nonlinear functions relating measurements to states,
and is the m 1 measurement error vector, whose m m
covariance matrix is assumed diagonal and denoted by R.
Therefore,
2
R =diag{12 , 22 , . . . , m
}
(2)
where i2 is the variance of measurement errors i .
Power system state estimation is formulated as the following weighted least squares problem:
min J(
x) =

1
t
[z h(
x)] R1 [z h(
x)]
2

(3)

where x
represents the vector of state estimates, and the
function to be minimized is the weighted sum of the squared
residuals.
2.2 Solution through Gauss-Newton Method
The solution of the minimization problem (3) through
the Gauss-Newton method leads to an iterative process in
which the following normal equation is solved in each iteration [9], [10], [11]


Ht R1 H x = Ht R1 z
(4)

where H is the Jacobian matrix of h(x) computed at a given

point xk and
k
z = z h(x ).

Solving Eq. (4) provides the vector x of increments to the

states, so that the updated state vector is obtained as
xk+1 = xk + x

17th Power Systems Computation Conference

The iterative procedure goes on until x becomes smaller

than a pre-specified tolerance.
The coefficient matrix on the left-hand side of Eq. (4) is
often referred to as gain matrix. As shown in [9], the inverse
of the gain matrix, that is,

1
P = Ht R1 H
(5)
can be interpreted as the covariance matrix of the estimation errors. Thus, it provides a measurement of confidence
on the accuracy of the state estimates [9].
2.3 Embedding A Priori State Information
There are practical situations in which, in addition to
measured data, some reliable knowledge on values of the
state variables is available prior to state estimation. This a
priori information, which we will denote by x
, may arise
in different forms, such as state estimates determined on the
basis of a previous set of measurements, or mean values
based on the expected range of values for the state variables. In addition, it is often possible to know additional
statistics concerning the a priori state vector, such as its covariance matrix. As remarked in [12], a priori information
contributes to the estimation process in a similar fashion as
the measured data. Also, the statistics required for modeling a priori information is the same statistics of the error
(
x x) [12]. If we denote by P the n n covariance matrix of x
, the weighted least-squares problem of equation (3)
can be extended to account for the a priori state information
by augmenting the objective function of problem (3) as [12]
min J(
x)

t
1
x)] R1 [z
2 [z h(
1
t 1
xx
) P (
xx
)
2 (

h(
x)] +

(6)

It can be shown that the optimality conditions for the augmented problem lead to the following extended normal
equation:
 t 1

H R H + P1 x = Ht R1 z + P1 x (7)

xk . Therefore, whenever a priori inwhere x = x

formation is considered, Eq. (7) replaces Eq. (4) in the process of solving the weighted least squares problem. We will
refer to this problem as state estimation in the presence of A
Priori State Information or, for short, APSI state estimation.
In practice, the values in x
are often assumed uncorrelated, in which case matrix P is diagonal. The i-th diagonal
entry of P will be denoted by
i2 , and corresponds to the
variance of the a priori information x
i . Ascribing infinite
values to
i2 , i = 1, . . . , n, means that nothing is known
in advance about the state variables. As a consequence,
P1 0 so that the second term on the right-hand side
of expression (6) would be null, and the problem reduces
itself to the conventional state estimation in which a priori
information is neglected.
Although employed to solve many practical problems,
the Gauss-Newton method is somewhat prone to numerical
ill-conditioning [13]. For this reason, algorithms that avoid
the explicit calculation of the gain matrix, such as the orthogonal methods discussed in next section, are often pre-

Stockholm Sweden - August 22-26, 2011

ferred when dealing with large real problems whose numerical conditioning cannot be easily assessed beforehand.
3 APSI State Estimation via Givens Rotations
3.1 Outline of three-multiplier Givens Rotations
Weighted least-squares problems can be solved by a
faster, three-multiplier (3-M) version of Givens rotations,
proposed in [14] to improve the computational performance
of the original four-multiplier rotations. To outline the
method, consider that one wishes to solve the linearized
weighted least-squares problem whose measurement model
is given by
z = Hx +
and the objective function is the linearized version of (3).
According to the orthogonal approach, a sequence of elementary rotations are applied to matrix H and vector z
(both previously scaled by matrix R1/2 ) in order to obtain
an upper triangular linear system of equations. If Q represents the matrix that stores the individual rotations, we have
[13]:





U c
21
H z
(8)
=
Q R
e
0
where U is an upper n n triangular matrix, and c and e
are n 1 and (m n) 1 vectors, respectively.
The 3-M version of Givens rotations is based on the decomposition of matrix U as [14],[13]:
1

U = D2 U

(9)

is a unit upper triangular matrix, that is, its diagowhere U

nal entries are all equal to 1, and D = diag{d1 , d2 , . . . , dn }.
Vector z is considered as an extra column of H, so that
c is also scaled in the transformation. The resulting scaled
vector is denoted by
c.
The artifice of scaling U as above provides significant
computational benefits, such as the reduction on the number of operations and the elimination of square-root computations during the factorization given by Eq. (8). In prac1
tice, D 2 is not actually required, so that only D needs to
be computed. Also, matrix Q does not have to be explicitly
formed.
Givens rotations are a row-processing scheme through
which every augmented row composed of a particular measurement zi and the corresponding row of H is processed
one at a time, until all information contained in the augmented row is transferred to matrix U. The cumulative effect of all elementary rotations required for that purpose is
represented by matrix Q in Eq. (8). After all measurements
have been processed, estimates
x are obtained by simply
solving the upper triangular system

U
x=
c

(10)

by back-substitution. The weighted sum of squared residuals is determined from e, as a by-product of the estimation
process.

17th Power Systems Computation Conference

3.2 Elementary Rotations and Interpretation of Scaling

Factors
We now focus the attention on the elementary 3-M
Givens rotations. The purpose is to show how this variant
of the rotations facilitates the implementation of APSI estimators through the proper interpretation of the triangular
matrix scaling factors introduced in Eq. (9). For conciseness, we omit the mathematical details and developments.
The interested reader is referred to [14] and [13].
We first consider the case in which no a priori information is taken into account. In this case, prior to the processing of any measurement matrix U and vector c in Eq. (8)
are both initialized as zero. From Eq. (9), this obviously
requires that all scaling factors for the rows of U be made
must always be unit upper triequal to zero, since matrix U
angular. Therefore, the entries di of diagonal matrix D are
also initialized as zero. As the measurements and the corresponding rows of H are individually processed, the values of di are repeatedly updated and, as a result, become
nonzero. In addition, the updating rule is such that di will
always be nonnegative [14].
The 3-M version of the rotations assumes that every augmented row formed by a measurement and the corresponding row of matrix
H is also scaled. Let us denote that scaling factor by w. In weighted least squares problems, the
value initially assigned to w is the weight attributed to the
corresponding measurement. From Eqs. (2) and (3) , we
conclude that, in the specific case of PSSE, the initial value
for the scaling factor for measurement zj should be the inverse of its variance, that is, wj = 1/j2 [13]. Measurement
scaling factors are also updated at each elementary rotation,
and their nonnegative values steadily decrease towards zero
as the information contained in the augmented row is grad
ually absorbed by the triangular matrix U.
Having established the statistical role of the measurement scaling factor w, an interpretation of the scaling fac rows is now in order. As shown in the
tors di of the U
sequel, this interpretation is intrinsically connected with the
way a priori state information can be taken into account in
the 3-M Givens framework. Analogy with w suggests that
(0)
the initial value di can also be seen as a weight, but in
this case assigned to the state variables (notice that there
are as many ds as states), and before any measurement is
processed. In other words, d(0) can be interpreted as the
weighting factor for the a priori information possibly avail(0)
able on state variable xi . In addition, the value for di must
be in agreement with the second term on the right-hand side
of Eq. (6), which establishes how a priori information is
taken into account in the estimation process. This leads to
the conclusion that
(0)

di

= 1/
i2

(11)

where
i2 is the variance of the a priori information on state
variable i.
As mentioned in the beginning of this subsection, the
practice in previous applications of the 3-M rotations to
PSSE (which neglect a priori state information) has been

Stockholm Sweden - August 22-26, 2011

to initialize
c = 0 and D = 0, which amounts to assuming
that U is initially a null triangular matrix [13]. This is consistent with the discussion in Subsection 2.3, since D = 0
actually means that nothing is known in advance about the
states, so that their a priori variances are infinite.
To sum up, from Eqs. (8), (9) and (10) we conclude
that previous information on the state variables can be easily considered in the 3-M version of Givens rotations framework by adopting the following simple steps:
i) To initialize vector
c as the available a priori state
information, that is,
(0)

=x

(12)

ii) To initialize di as given by Eq. (11).

Since the original formulation of 3-M Givens rotations
already makes provision for those variables, no extra computational cost is actually incurred when a priori state information is taken into account.
4 Proposed Strategy to Enhance PSSE by Processing
PMU Data
As previously mentioned in this paper, the proposed approach to take advantage of PMU data for real-time modeling recognizes the difficulties to fully merge information obtained from PMUs and data gathered by traditional SCADA
systems. In addition to technological aspects, such as differences in instrumentation and distinct data gathering rates,
relevant arguments from the software standpoint can also be
put forward in favor of separating the processing of PMU
and SCADA measurements, as already mentioned. This of
course raises the challenge as how to combine the information provided by both sources in the most beneficial way, in
terms of both reliability and accuracy.
In this paper, the proposed solution to meet such challenge consists of a two-stage scheme, in which each stage
processes a single type of data (either SCADA or PMU).
Figure 1 illustrates the suggested methodology. The first
stage consists of a conventional state estimator which processes SCADA measurements only. There is no restrictions
concerning the algorithm employed at this stage, so that any
existing state estimator can be used. This estimator should
provide the SCADA-based state estimates, xS , as well as
the corresponding covariance matrix of estimation errors,
PS , as given by Eq. (5). However, as it will become apparent in the sequel, only the diagonal elements of PS are actually needed. Ideally, those values would be directly available as a byproduct of state estimation, since they are also
required to compute the normalized residuals for bad data
processing [10]. However, if that is not the case for a given
state estimator, the diagonal entries of PS can be easily rebuilt from the Jacobian matrix H and the measurement error covariance matrix R. The sparse inverse matrix method
provides an efficient means to calculate only the required
elements of PS [15].
In the second stage depicted in Fig. 1, the APSI orthogonal state estimator described in Section 3 is employed.

17th Power Systems Computation Conference

Measurements processed by that estimator are restricted to

those gathered from available PMUs, but the SCADA-based
estimates determined in the first stage also take part in the
estimation process. This is ensured by treating the latter as
a priori state information to the APSI estimator. To accommodate all that, the procedure discussed in Subsection 3.2
are employed. We thus make use of Eqs. (11) and (12)
to define the a priori information and respective variances
applicable to this particular problem. From previous arguments, we can easily conclude that:

c(0)
(0)
di

=
=

x
S
1/PS,ii

(13)

As the PMU measurements are sequentially processed

by the APSI estimator, the a priori state information determined in the first stage, that is, the former SCADA-based
estimates, undergo successive changes so that they are repeatedly updated by the Givens rotations row-processing
mechanism. Since PMU measurements are usually considered more accurate than SCADA data, it is expected that
the quality of the state estimates will be enhanced by the
processing in the second estimation level.

Figure 1: Proposed 2-stage APSI state estimator.

5 Additional Remarks
The following remarks are intended to complement and
exploit some features of the two-stage estimator described
in the previous section:
Although observability is assumed at the first-level
SCADA-based estimation process, the network does not
have to be observable with respect to the measurement
subset composed by PMU measurements only. In fact, if
the network is observable at the SCADA level, than the
condition to obtain an state estimation solution is ensured
by the a priori information scheme described in Section
4. In other words, observability at the SCADA-based
level ensures the observability for the overall estimation
process. This is a convenient result, since in most existing power systems the number of PMU installations are
not enough as yet to cover the whole electric network;

Stockholm Sweden - August 22-26, 2011

 On the other hand, the existence of PMU measurements

in a given part of the electric network tends to enhance the
quality of state estimation results, due to the superior accuracy of those measurements with respect to SCADAs.
A convenient form to understand that is to think of the
PMU data as providing an improved estimation finishing coat on top of the primer already provided by the
a priori SCADA-based state information;
Another benefit of processing PMU measurement in the
second stage is to improve the degree of redundancy in
those parts of the network where phasor measurements
are available. As illustrated through simulations in the
next section, this could be very helpful in filtering out
the effects of bad data contaminating conventional measurements, particularly when the SCADA measurement
redundancy in the same region is poor;
In case of loss of all PMU information, the proposed estimator would still provide the state estimates based on
the SCADA measurements, that is, the conventional state
estimates;
The state estimation model employed in this paper for the
second stage of Fig. 1 is based on polar coordinates and
is nonlinear. However, there is no apparent theoretical
reason to impede the adoption of another formulation of
state estimation equations leading to a linear estimation
model. This objective will be pursued in the next stages
of the research on this topic;
Likewise, further investigation on the potential benefits
of the proposed method concerning other issues related
to state estimation is also needed. This is the case of bad
data processing at the second estimation level in Fig. 1.
6

presents the number of measurements of each type to be

processed by both conventional and APSI estimators. By
simple inspection, we can easily conclude that the network
is not PMU-observable, since only 9 phasor measurements
are available. On the other hand, the system is SCADAobservable, despite the presence of some critical measurements in the corresponding metering scheme.
Two test cases are performed with the IEEE 14-bus system, as follows:
Case A-1: The base case, in which all SCADA measurements processed at the first stage are all free of gross
errors, thus providing error-free a priori information
for the APSI Estimator;
Case A-2: The voltage magnitude SCADA measurement at
bus 10 is considered contaminated with a gross error
of size equal to 10 times the measurements standard
deviations. As a consequence, the a priori state information transferred to the APSI estimator is biased by
the simulated error.

Simulation Results

This section illustrates the application of the proposed

APSI orthogonal estimator through several case studies conducted with the IEEE 14-bus, 57-bus and 118-bus test systems. A conventional SCADA-based state estimator is employed as a reference to evaluate the performance of the proposed approach for all cases. Both estimators have been
programmed in Fortran and run on a 2.9 GHz, 4.0 GB of
RAM Intel Core2 Duo computer. The measurement accuracy levels employed to perform the tests are 1% for
SCADA and 0, 1% for PMU measurements.
To assess the performance of the estimators we make
use of the voltage metric defined in [16]:

1



2 2
X


 true

M accv = V error =
V jest  (14)
V j
2

where V jtrue and V jest are the true and estimated complex
phasor voltage at the j-th bus, respectively. In addition, the
mean and standard deviation of the bus voltage magnitude
and phase estimation errors are also employed as performance indices in some of the test cases.
6.1 IEEE 14-bus system
The metering scheme for the IEEE 14-bus system is indicated on the one-line diagram shown in Figure 2. Table 1

17th Power Systems Computation Conference

Figure 2: IEEE 14-bus system with SCADA and PMU measurements

P
Q
|V |
t
u
SCADA measurements
7
6
8
10 9
PMS measurements
0
0
0
0
0
Table 1: Number of SCADA and PMU measurements for
system.

V
0
8
IEEE

I
0
1
14-bus

x 10

APSI Estimator
SCADAbased
estimator
Unobservable states
with PMU meas. only

10

12

14

Figure 3: Voltage magnitude error for IEEE 14-bus system. Case A-1.

Stockholm Sweden - August 22-26, 2011

x 10

APSI Estimator
SCADAbased estimator
Unobservable states
with PMU meas. only

3.5
3
2.5
2
1.5
1
0.5
0

10

12

14

Figure 4: Voltage angle error for IEEE 14-bus system. Case A-1.

The results for Case A-1 are presented in Figures 3 and

4, which show the estimation errors for voltage magnitudes
and angles at each bus of the network. The solid and dashed
lines drawn in both figures have the sole purpose of facilitating the identification of the APSI and SCADA-based estimates. Some interesting conclusions can be extracted from
both plots, namely:
The results of the APSI estimator show that the estimates of state variables which are not PMU-observable
are equal to the corresponding a priori estimates provided
by the SCADA-based estimator. This is indicated in both
figures by the coincidence of APSI and SCADA-based
estimation errors for those buses at which no PMU is installed (the exception is the voltage phase at bus 1, where
the identical error values are due to the definition of that
variable as the angular reference for both estimators);
On the other hand, the APSI estimation errors for the
buses at which there is at least one adjacent phasor measurement are considerably reduced, as a consequence of
the enhanced quality of the PMU data. These results also
confirm the effectiveness of the data weighting strategy
adopted in the proposed APSI formulation, since the a
priori information in this case are assigned much less
weight than the PMU measurements.
Case A-2 differs from the previous one in that a bad
data is simulated on the voltage magnitude SCADA measurement at bus 10. As it turns out, that measurement is
critical, what magnifies the influence of the gross error on
the SCADA-based estimate for that state variable, as depicted in Fig. 5 (for case A-2, only the voltage magnitude
errors are shown). In fact, the estimate for voltage magnitude at bus 10 is exactly equal to the value of the erroneous
measurement. Since bad data on critical measurements are
undetectable [10], this is a particularly delicate situation.
However, Fig. 5 also shows that, even starting from an
a priori information contaminated by the gross error, the
APSI estimator is able to filter the error out, since there
is a reliable phasor measurement (current measurement on
branch 9-10) adjacent to the affected state variables. Results for the voltage metric of Eq. (14) for the same case are
given in Table 2 and corroborate the previous analysis.
A situation similar to case A-2 but with no PMU measurement adjacent to the faulty measurement has also been

17th Power Systems Computation Conference

evaluated. In this case, the measurements processed at the

second stage add no information to enhance the estimate
for state variable |V10 | . Consequently, the final estimate for
that variable equals the erroneous value originated from the
SCADA-based estimator.
Additional results considering the presence of bad data
contaminating members of critical sets of the SCADA metering scheme have also been simulated, but the details are
omitted of this paper due to space limitations. As in case A2, it has been observed that the APSI estimator is also able
to filter out the gross errors for those cases, provided that
the region where the errors occur is properly monitored by
PMU measurements.
0.14

APSI Estimator
SCADAbased estimator

0.12
0.1
0.08
0.06
0.04
0.02
0

10

12

14

Figure 5: Voltage magnitude error for IEEE 14-bus system. Case A-2.
Voltage Metric
Estimator
Case A-1
Case A-2
SCADA-based
1.7658e-002 1.5292e-001
APSI
1.0141e-002 1.0141e-002
Table 2: IEEE 14-bus performance index

6.2 IEEE 57-bus and 118-bus systems

The proposed APSI estimator has been also applied
to the IEEE 57-bus and 118-bus test networks. The metering schemes employed for both systems are presented
in Tables 3 and 5, and exhibit some common characteristics: all bus injection and branch flow power measurements
are taken in active/reactive pairs and branch current measurements are considered at both estimation stages, either
as magnitude measurements (SCADA) or phasor measurements (PMUs). Both systems are SCADA-observable, but
PMUs are not widely deployed throughout neither network,
so that they are not PMU-observable. Finally, the cases reported in this paper do not include any bad data simulation.
The results for the 57-bus and 118-bus systems are presented in Tables 4 and 6, respectively. They include values
of the voltage metric defined in Eq. (14), as well as bus voltage magnitude and angle mean and standard deviation values, computed for both the SCADA and APSI state estimators. As expected, voltage metric and error mean values for
the APSI scheme are much lower than those obtained from
the conventional SCADA estimator. For phase angles, the
observed reduction is larger than 55% in both systems. Error standard deviation values are of the same order of magnitude for both estimators, but it is possible that the APSI
values become larger than SCADAs. This is certainly due

Stockholm Sweden - August 22-26, 2011

to the non-uniform distribution of the PMU measurements

throughout both networks, that is, the coexistence of zones
with large concentration of PMUs and zones with no installled PMUs, leading to a larger estimation error spreading
around the error mean value in the APSI case.
P,Q
|V |
|I|
t,u
V
I
SCADA measurements
39
36
39
52
0
0
PMS measurements
0
0
0
0
30 20
Table 3: Number of SCADA and PMU measurements for IEEE 57-bus
system.
Voltage
Error Mean
Estimator
Metric
|V |

SCADA-based
4.42e-3
2.70e-4 4.82e-4
APSI
2.60e-3
1.88e-4 1.77e-4
Table 4: IEEE 57-bus performance index

Error Std Dev

|V |

2.30e-4 1.70e-4
1.45e-4 2.14e-4

P,Q
|V |
|I|
t,u
SCADA measurements
72
67
45
116
PMS measurements
0
0
0
0
Table 5: Number of SCADA and PMU measurements for
system.
Voltage
Error Mean
Estimator
Metric
|V |

SCADA-based
6.69e-3
2.78e-4 4.27e-4
APSI
4.47e-3
1.99e-4 1.91e-4
Table 6: IEEE 118-bus performance index

V
0
63
IEEE

I
0
65
118-bus

Error Std Dev

|V |

1.24e-4 3.40e-4
1.81e-4 2.59e-4

7 Conclusions
A two-stage state estimator to allow the incorporation
of PMU measurements into power system real-time modeling without the risk of disrupting existing SCADA-based
estimators is proposed. Any existing conventional state estimator can be employed in the first stage. The resulting estimates are treated as a priori state information by the estimator at the second level, which processes only PMU measurements. The latter, referred to as APSI estimator, is based on
an specialized orthogonal row-processing algorithm which
exhibits as one of its properties the capability of processing a priori information at virtually no extra computational
cost. An attractive feature of the proposed scheme is that
PMU-observability is not required, as long as the system is
SCADA-observable.
The results presented in the paper show that the APSI
estimator is able to take full advantage of the superior quality of PMU data in order to enhance the a priori information generated by the SCADA estimator. This applies even
to cases in which the SCADA estimates are obtained in the
presence of bad data, provided that reliable PMU measurements are available in the vicinity of the SCADA gross measurements.
The proposed estimator architecture is such that an EMS
equipped with a conventional state estimator and access
to PMU measurements taken on the SCADA-monitored
network could easily update its real-modeling capability
through the APSI scheme. The amount of information
needed to be transferred between both modules is not large,
since it consists of the SCADA state estimates and corresponding estimation error variances.
Finally, a number of issues still deserve further research
efforts, such as bad data processing at the second level and

17th Power Systems Computation Conference

Powered by TCPDF (www.tcpdf.org)

the measurement model employed by the APSI estimator,

which is presently based on polar coordinates and nonlinear. Since there is no theoretical reason to prevent the adoption of other modeling frameworks, the use of a linear measurement model seems to be feasible. Those topics will be
pursued in the next stages of this research.
REFERENCES
[1] A. Bose. Smart transmission grid applications and their supporting infrastructure. Smart Grid, IEEE Transactions on,
1(1):11 19, 2010.
[2] Jun Zhu and A. Abur. Effect of phasor measurements on the
choice of reference bus for state estimation. In Power Engineering Society General Meeting, 2007. IEEE, pages 1 5,
2007.
[3] Jun Zhu and A. Abur. Bad data identification when using
phasor measurements. In 2007 IEEE Power Tech, pages 1676
1681, Lausanne, Switzerland, 2007.
[4] R.C. Leites and A. Simoes Costa. An Orthogonal State Estimator with Phasor Measurement Processing Capability. In
3rd. Brazilian Symposium of Electric Systems, Belem, Brazil,
May 2010. (in Portuguese).
[5] Ming Zhou, V.A. Centeno, J.S. Thorp, and A.G. Phadke. An
alternative for including phasor measurements in state estimators. Power Systems, IEEE Transactions on, 21(4):1930
1937, 2006.
[6] R.F. Nuqui and A.G. Phadke. Hybrid linear state estimation utilizing synchronized phasor measurements. In 2007
IEEE Power Tech, pages 1665 1669, Lausanne, Switzerland, 2007.
[7] A. Simoes Costa, E. Lourenco, and F. Vieira. topology error
identification for orthogonal estimators considering a priori
state information. In Proceedings of the Power System
Computation Conference, Li`ege, Belgium, Ago 2005.
[8] A. Simoes Costa, R.S. Salgado, and P. Haas. Trust region
optimization methods via Givens rotations applied to power
system state estimation. In E. Castronuovo, editor, Optimization Advances in Electric Power Systems, chapter 2, pages
2751. Nova Science Publishers, Inc., New York, USA,
2008.
[9] F. C. Schweppe and J. Wildes. Power System Static-State
Estimation, Part I: Exact Model. IEEE Trans. on Power Apparatus and Systems, 3(4):120125, Jan. 1970.
[10] A. Monticelli. State Estimation in Electric Power Systems:
A Generalized Approach. Kluwer Academic Publishers,
1999.
[11] A. Abur and A. Gomez Exposito. Power System State Estimation - Theory and Implementation. Marcel Dekker, 2004.
[12] P. Swerling. Modern State Estimation Methods from the
Viewpoint of the Method of Least Squares. IEEE Trans. on
Automatic Control, 16(6):707719, December 1971.
[13] A. J. A. Simoes Costa and V. H. Quintana. An Orthogonal Row Processing Algorithm for Power System Sequential
State Estimation. In IEEE PES Winter Meeting, Atlanta USA, Jan 1981.
[14] W.M. Gentleman. Least Squares Computations by Givens
Transformations Without Square Roots. Journal of the Inst.
Math. Applics., pages 329336, 1973.
[15] F. Broussole. State estimation in power systems: Detecting
bad data through the sparse inverse matrix method. IEEE
Transactions on Power Apparatus and Systems, 97, Mai-Jun
1978.
[16] KEMA. Metrics for Determining the Impact of Phasor Measurements on Power System State Estimation. Eastern Interconnection Phasor Project, March 2006.

Stockholm Sweden - August 22-26, 2011