Documente Academic
Documente Profesional
Documente Cultură
Table of Contents
from
A. P. Sakis Meliopoulos
Page 2
Chapter 7
Operating State Estimation
7.1 Introduction
Effective control and operation of electric power systems requires accurate and reliable
knowledge of the system model and the operating state of the system in real time. For
this purpose, modern power systems are equipped with an extensive data acquisition
system. Local analog and status quantities, such as voltage magnitude, real and reactive
power flows, loads, status of breakers (open/close) etc., are measured and transmitted to a
central location. The measurements are simple, requiring simple instrumentation.
Typical analog measurements are: (a) voltage magnitudes, (b) real and reactive power
flows and (c) current magnitude measurements. Recent technology based on GPS
(Global Positioning System) has made it possible to measure voltage phase angles as
well. Typical status measurements are: (a) breaker status, (b) disconnect switch status,
etc. All measurements are taken every one to several seconds. They are transmitted to a
central location (the Energy Management System (EMS) or the Energy Control Center)
where they are processed to yield the operating state of the system. The process consists
of two analysis problems: (a) determination of network topology, and (b) determination
of operating state. The network topology is constructed from the status of breakers and
disconnect switches. The operating state of the system is constructed from analog
measurements by means of two distinct computational procedures: (a) on-line power flow
and (b) state estimation.
On-line power flows utilize a subset of available measurements which are enough to
define the power flow problem. Solution of the power flow problem yields the operating
state of the system. Because measurements are usually corrupted with error (resulting
from PT or CT inaccuracies, instrument error, transmission error, etc.), the error is
directly transmitted to the computed operating state. It is also possible that one of the
measurements may include a large error (gross error - due to meter malfunctioning or
communication errors) resulting in a non-solvable power flow problem or in a solution
that may be quite different than the actual operating condition of the system. In a real
time environment, it is important to have the ability to identify wrong measurements or
errors in measurements (bad data). For this purpose, it is necessary to take advantage of
redundant measurements. The redundant measurements are utilized to compute the best
estimate of the operating state of the system with statistical methods. The computational
procedure is called state estimation. As we shall discuss, state estimation provides the
mechanism to: (a) determine whether the system state can be computed from existing
data (observability); (b) filter out usual measurement errors and, therefore, compute the
system state with minimum error; (c) identify and reject bad data; and (d) determine the
degree of confidence on the estimated state of the system. The conceptual view of the
Page 3
process is illustrated in Figure 7.1. The two approaches, i.e. on-line power flow and state
estimation, are identified. In addition other applications are possible using the collected
data. These applications are parameter estimation and remote calibration of the meters.
The constituent parts of this process will be discussed next.
Bad Data
Identification
& Rejection
Observability
Analysis
Status
Data
Topology
Processor
Analog
Data
On-Line
Power
Flow
Solution
State
Estimation
Best Estimate of
System State
& Model
System State
Parameter
Estimation
Remote
Calibration
Figure 7.1 Conceptual View of Real Time Power System Modeling and
State Estimation
The objective of this chapter is to introduce the models involved in the state estimation
problem and to discuss its various extensions and applications. Because normally there
are redundant measurements, it is possible to use the measurements for the purpose of
determining or improving the system model parameters or to remotely calibrate the
meters.
Page 4
the past it was customary to have separate communication channels for the Supervisory
Control and Data Acquisition. Today, however it is a unified system sharing a common
two way communication channels that may consist of several physical layers.
Communications are integrated in the remote terminal unit (RTU) which manages data
collection, control functions and communication with a master station. The master
station has multiple communication channels to remote terminal units. Many times a
dedicated channel is assigned to each remote terminal unit. In other cases, there are less
channels than remote terminal units requiring more than one remote to share a channel.
Analog data is scanned periodically, typically every one second to a few seconds. Each
scan is triggered by the system control center at the prescribed interval by using a request
to all remote stations to send in data.
The amount of data collected and transmitted is very large for typical power systems.
These data is transported via communication channels. In order to minimize
communication traffic some form of data compression is utilized. For example, for status
data one can send only changes of status data. This approach minimizes amount of data
transfer and the amount of processing needed at the master station. Analog data can be
also compressed with a number of methods.
Independently of system configuration, SCADA system manufacturer, communication
software and computer configuration, the end result of the SCADA system function will
be the collection of a set of system data every sampling period. The data consists of:
Breaker status
Disconnect switch status
Transformer tap setting
MW flow measurements
MVAR flow measurements
Voltage magnitude (kV) measurements
Current magnitude (kA) measurements
phase angle difference measurements
etc.
Page 5
G1
G2
MW Flow Measurement
MVAR Flow Measurement
kV Measurement
Disconnect Switch Status
Breaker Status
RTU
Communication
Link with Control
Center
Data
RTU
Commands
Master
Station
Page 6
The data are utilized to form the system model (Network Configurator) and to estimate
the system operating state (state estimation). The next sections describe these
applications.
AutoBank
500kV/230kV
G1
AutoBank
500kV/230kV
G2
Page 7
SG1
SG2
Figure 7.3b Bus Oriented Network Model of the System of Figure 7.3a
-j15.0
Page 8
zi = hi (x )
(7.1)
where x is the system state and hi is a function specific to the measured quantity zi.
Assume that m measurements are taken. Then:
z = h( x )
(7.2)
where
x is the system state - an n x 1 vector
z is a vector of measured quantities - an m x 1 vector
h is a vector function - an m x 1 vector function.
Typically more measurements are taken than the number of state variables to be
determined, i.e. m>n. In this case, the set of Equations (7.2) represents an
overdetermined set of nonlinear equations in real variables.
In general an overdetermined set of equations, such as (7.2) does not have a solution for
x. Only if the system model expressed with the functions h(x) is exact and the
measurements z are extremely accurate, then equations (7.2) have a unique solution for
x. This of course is unlikely in a real system. Yet it is possible to obtain a solution for x
via a procedure known as the state estimation. This procedure will be first introduced by
an example and then it will be discussed in more detail.
Example E7.1: Consider the simplified power system of Figure E7.1. The following
measurements are taken (all in p.u.):
z1 = V1 = 1.0,
z3 = P21 = -1.56,
z2 = V2 = 0.99
z4 = Q21 = 0.01
~
V1 =V1e j0
-j15.0
~
V2 = V2e j2
Figure E7.1
Solution: The state vector x is:
Page 9
V1
x = V2
2
The vector function h(x) is given as:
h1 ( x ) = V1
h2 ( x) = V2
h3 ( x ) = 15.0V1V2 sin 2
h4 ( x ) = 15.0V22 15.0V1V2 cos 2
(7.3)
where:
b
h(x)
x
m>n
is a known m x 1 vector
is an m vector of known function
is the unknown n x 1 vector
(7.4)
where:
b
H
x
m > n.
Page 10
is a known m x 1 vector
is a known m x n matrix
is the unknown n x 1 vector
or
r = h ( x ) b (nonlinear case)
(7.5)
r = Hx b (linear case)
(7.6)
and compute a solution x which will minimize the residual vector in some sense. For this
problem, there are three approaches: (a) the least squares solution, (b) the least absolute
deviation solution(L1 approach), and (c) the Chebyshev or min-max solution(L
approach). These methods will be briefly discussed next.
An Alternative Introduction of the State Estimation Problem: An alternative way to
introduce the state estimation problem is by considering the hardware used for obtaining
the measurements. Any metering device comprises an instrumentation channel, a
transducer and an A/D converter (in recent systems the transducers are omitted since
they are not necessary). The intrumentation channel comprises instrument transformers
(Potential Transformers (PT), Current Transformers (CT), Optical Transformers, etc.),
interconnecting cables and possibly attenuators. The transducer may be an analog
device that converts the input signal into a DC output signal proportional to the quantity
measured. The A/D converter samples the signal and convertes it into a ditital form. The
signal can be the output of the transducer or it can be the output of the instrumentation
channel. In the latter case, an intelligent device is required to extract the appropriate
information, i.e. rms value of the voltage waveform, real power flow, etc. A pictorial
view of this arrangement is shown in Figures 7.5a and 7.5b.
PT
Control Cable
Transducer
Output
0-10 V DC
Page 11
PT
Control Cable
A/D
Conversion
Digital
Interface
E{ i } = 0
E{ i2 }= i2
E{ i j } = 0, i j
(7.7)
(7.8)
(7.9)
Consider a specific measurement bi. This measurement is related to the state of the
system via a known function hi(x). We have named this function the model of the
measurment. We postulate that the difference between the model and the measurement
is the rror of the measurement, i.e.:
i = hi ( x ) bi
(7.10)
It should be recognized that the measurement error is the same as the measurement
residual we have introduced already. The difference is that now we can associate
statistics this residual. In subsequent paragraphs we shall use equations (7.5) and (7.10)
interchangeably and will assume that the residuals r or the errors h have the statistics
stated with equations (7.7), (7.8) and (7.9).
Page 12
Minimize
J = ri 2 = r T r = T
(7.11)
i =1
A variation of this method is the weighted least squares method which minimizes the sum
of the weighted squares of the components of the residual vector r or the vector of the
measurement errors . Mathematically, this is expressed as follows:
m
Minimize
J = wi ri 2 = r T Wr = T W
(7.12)
i =1
where:
wi : the weight for the residual ri
W : a diagonal matrix, the diagonal elements being the weights wi.
A most usual case of weighted least squares is defined as follows. We postulate that we
want to compute the state of the system that minimizes the sum of the squares of the
normalized measurement errors, si. The normalized measurment error is defined with:
si =
i
i
(7.13)
Minimize
= s i2 = T W
i =1
(7.14)
where:
1 1
1
W = diag 2 , 2 , ..., 2
m
1 2
Note that this is a weighted least squares formulation with the weights defined as the
inverse of the squared standard deviations. It should be also noted that the formulation in
terms of equation (7.11) is equivalent to assuming that all standard deviations of the error
of all measurements are equal.
In subsequent paragraphs we will consider the weighted least squares approach.
Page 13
Using the nonlinear and linear model equations, the nonlinear and linear state estimation
problem is expressed as follows:
Minimize J = (h( x ) b) T W (h( x ) b) , for the nonlinear case
(7.15)
(7.16)
The unknown vector x is obtained from the solution of the necessary conditions, which in
matrix notation are expressed as follows:
dJ
=0
dx
(7.17)
The above problem is fisrt solved for the linear case and then for the nonlinear case.
Linear Case: Direct differentiation of equation (7.16) with respect to x, we obtain:
dJ
d
= [( Hx b) T W ( Hx b)] = 2 H T W ( Hx b) = 0
dx dx
(7.18)
x = H T WH
H T Wb
(7.19)
Equation (7.19) provides the solution to the linear estimation problem (7.16).
Nonlinear Case: To obtain the solution to the nonlinear estimation problem (7.15),
assume that an initial guess of the vector x0 is known. The nonlinear model equations
(7.5) are linearized around the point x0 yielding:
r = = h( x 0 ) +
h( x )
|
( x x 0 ) + h.o.t. b
x x = x
0
Where h.o.t. denotes higher order terms. Assuming that the vector x0 is very close to the
solution, then the higher order terms (h.o.t.) are negligibly small and are omitted from
above equation, yielding:
r = =
Let
h( x )
| x = x ( x x 0 ) + h( x 0 ) b
x
0
(7.20)
h( x )
| x = x = H , and b ' = h( x 0 ) + Hx 0 + b
x
0
Page 14
Observe that the vector b ' = h( x 0 ) + Hx 0 + b is known (or computable). Now equation
(7.20) becomes:
r = = Hx b '
Now the problem is identical to the linear estimation problem. Thus, the solution is:
x = H T WH
H T Wb '
x = H T WH
H T W ( Hx 0 h( x 0 ) + b) = x 0 H T WH
H T W ( h( x 0 ) b)
x +1 = x ( H T WH ) 1 H T W (h( x ) b)
(7.21)
x 0 = [1.0 0.99 0.0] and least squares estimation. Assume that all weights are equal to
1.0.
T
Solution: The problem will be solved with the iterative algorithm (7.21). Since the
weights are all 1.0 the weight matrix W is the identity matrix. The objective function is:
J = (V1 1.0) + (V2 0.99) + (15V1V2 sin 2 + 1.56) + (15V22 15V1V2 cos 2 0.01)
2
Min J = wi (hi ( x ) z i )
i =1
Page 15
0
H =
15V 2 sin 2
15V 2 cos 2
0
1
15V1 sin 2
30V 2 15V1 cos 2
15V1V 2 cos 2
15V1V 2 sin
0
0
0.0
0.0
0
0
r = h( x ) b =
1.56
0.1585
0.0
0.0
1.0
0.0
1.0
0.0
H (x0 ) =
0.0 14.85
0.0
0.0
221.5225 218.295
H H = 218.295
217.09
0.0
0.0
220.5225
0.0
T
2.3537
H r = 2.33
23.166
T
0.00538
( H H ) H r = 0.00532
0.10505
T
0.0 0.10505 0.10505
1
2nd Iteration
Page 16
0.0054
0.0053
r 1 = h( x 1 ) b =
0.0030
0.0823
0.0
0.0
1.0
0.0
1.0
0.0
1
H (x ) =
1.5655 1.5644 14.767
220.589 0.000647
223.89
H H = 220.589
229.114
46.4911
220.489
0.000647 46.4911
T
1.2320
H r = 1.2369
0.0838
T
(H H )
0.00281
( H T H ) 1 H T r 1 = 0.00273
0.000196
0.9974
x = 0.9926
0.1052
2
This completes the solution. It is expedient to compute the residuals (or measurement
errors) using the above computed state:
r1 = V1 1.0 = 0.0026
r2 = V 2 0.99 = 0.0026
Page 17
Note that in this case the residuals or measurement errors are very small.
Minimize J = | ri |
i =1
A variation of this method is the weighted least absolute deviation method which
minimizes the sum of the weighted sum of the absolute deviations of the components of
the residual vector r. Mathematically, this is expressed as follows:
m
Minimize J = | wi ri |
i =1
The weights can be selected as discussed in the previous method. Again we will consider
the solution of the weighted least absolute deviation approach.
This requirement translates into the following optimization problem:
m
Minimize J = | wi ri |
i =1
Subject to: ri = hi ( x ) bi ,
i = 1, 2,...., m
Minimize J = | wi ri |
i =1
m
i = 1, 2,...., m
j =1
r = h( x ) b H ( x x 0 ) + h( x 0 ) b = Hx b '
Page 18
where:
x = x x 0
b ' = b h( x 0 )
Minimize J = | wi ri |
i =1
m
i = 1, 2,......, m
j =1
r = r+ r,
x = x + x ,
r+,r 0
x+, x 0
Minimize
Subject to:
J = w i (ri+ + ri )
i =1
r r H ( x + x ) = b '
r+ ,r, x+ , x 0
Minimize
Subject to
J = (ri+ + ri )
+
1
+
2
+
3
i =1
r r V1+ + V1 = 0.0
r r V2+ + V2 = 0.0
r r 14.85 2+ + 14.85 2 = 1.56
Page 19
by definition, there was no need to introduce + and variables for the voltage
magnitudes.
Upon solution of the above linear program we obtain the followings:
V1 = 0.01067
2 = 0.10505
r1 = 0.01067
and all other rs equal to zero
Thus, the new state variables and residuals are:
V1 = 0.98933
V 2 = 0.99
2 = 0.10505
r1 = r1+ r1 = 0.01067
r2 = r2+ r2 = 0.0
r3 = r3+ r3 = 0.0
r4 = r4+ r4 = 0.0
Note that only one residual is nonzero. In general, the LAV solution will result in m-n
maximum number of non-zero residuals. In this case m=4 and n=3, thus 4-3=1 residuals
will be nonzero.
On the other hand, using the nonlinear model equations and the above computed state,
the residuals are computed to be:
r1 = V1 1.0 = 0.01067
r2 = V2 0.99 = 0.0
Minimize
J = (ri+ + ri )
i =1
Subject to
+
1
+
2
r r1 V1+ + V1 = 0.01067
r r2 V2+ + V2 = 0.0
Page 20
V1 =0.00559
V2 = 0.0
2 = -0.00074
Vl = 0.9949
V2 = 0.99
2 = -0.1058
r1 = r1+ r1 = 0.005112
r2 = r2+ r2 = 0.0
r3 = r3+ r3 = 0.0
r4 = r4+ r4 = 0.0
Using the above computed state and the nonlinear model equations, the actual residuals
are computed to be:
r1 = V1 1.0 = 0.0051
r2 = V 2 0.99 = 0.0
This completes the solution. Note again the residuals (or measurement errors) are low in
this case.
r * = max( r1 , r2 ,...., rn
where ri is the i-th component of the residual vector r. This requirement translates into
the following optimization problem:
Minimize
Subject to
J = r*
wi hi ( x ) bi r * , i = 1,2,..., m
Page 21
r*
Subject to
wi
h
j =1
ij
x j bi r * , i = 1,2,..., m
r*
wi (hi ( x ) bi ) r * , i = 1,2,..., m
wi (hi ( x ) bi ) r * , i = 1,2,..., m
Upon linearization of the residuals:
h( x ) b H ( x x 0 ) + h( x 0 ) b = Hx b '
where:
x = x x 0
b ' = b h( x 0 )
Upon substitution:
Minimize
Subject to
r*
W ( Hx R * ) b '
W ( Hx + R * ) b '
where,
r *
*
r
*
R =
M
*
r
an m x 1 vector.
x = x + x ,
Page 22
x+, x 0
and introduce the slack (y) and surplus variables (z). Then,
Minimize
r*
Subject to
W ( Hx + Hx R * ) + y = b '
W ( Hx + Hx + R * ) z = b '
x+, x-, r*, y, z 0
Above problem is a standard linear problem. Upon solution, the state estimation problem
is solved. An example will demonstrate the method.
Example E7.4: Consider the problem stated in Example E7.2. Again assume the
weights to be equal to 1.0. Compute the solution x T = [V1 V2 2 ] , using the
Minimize
r*
Subject to
V1 1.0 r *
V2 0.99 r *
15.0V1V2 sin 2 + 1.56 r *
15.0V22 15.0V1V2 cos 2 0.01 r *
Upon removal of absolute values, linearization of the resulting equations around the
operating point x T = [V1 V2 2 ] = [1.0 0.99 0.0] and introduction of slack and
surplus variables, the above problem is translated into:
r1 = V1 1.0 = 0.0
r2 = V 2 0.99 = 0.0
Minimize
r*
Subject to:
V1+ V1 r * + y1 = 0.0
V1+ V1 + r * z1 = 0.0
Page 23
V2+ V2 r * + y 2 = 0.0
V2+ V2 + r * z 2 = 0.0
0.0V1+ 0.0V1 + 0.0V2+ 0.0V2 + 14.85 2+ 14.85 2 r * + y 3 = 1.56
0.0V1+ 0.0V1 + 0.0V2+ 0.0V2 + 14.85 2+ 14.85 2 + r * z 3 = 1.56
r* 0, y, z 0
Upon solution of above linear program, the following system state is computed:
V1+ V1 r * + y1 = 0.0052
V1+ V1 + r * z1 = 0.0052
V2+ V2 r * + y 2 = 0.0052
V2+ V2 + r * z 2 = 0.0052
Page 24
y, z 0
Upon solution of above problem, the following system state and residuals are obtained:
V1 =0.0027
V2 = - 0.0027
2 = - 0.0007
Vl = 0.9975
V2 = 0.9925
2 = - 0.1054
rl = | -r* + y1 | = 0.0025
r2= | -r* + y2 | = 0.0025
r3 = | -r* + y3 | = 0.0025
r4 = | -r* + y4 | = 0.0025
Using the above system state and the nonlinear model, the following residuals are
obtained:
r1 = V1 1.0 = 0.0025
r2 = V 2 0.99 = 0.0025
LS
0.9974
0.9926
-0.1052
-0.0026
LAV
0.9949
0.99
-0.1058
0.0051
min-max
0.9975
0.9925
-0.1054
0.0025
Page 25
r2
r3
r4
0.0026
0.000629
0.000631
0.0
0.0002
0.00015
0.0025
-0.0023
-0.002027
hi ( x ) bi
We have assumed that the normalized errors are Gaussian distributed with standard
deviation 1.0 and zero cross correlation.
Given the meter accuracy defined above, two problems can be defined as follows: (a)
what is the probability that all data are located within expected bounds (Goodness of Fit)
and (b) what is the accuracy of the computed solution? These problems will be
addressed next.
Page 26
The goodness of fit is defined as the probability that the distribution of the
measuremenmt errors are within the expected bounds. This probability is computed as
follows. Assume that the state estimate x$ has been computed with the least square
approach. Consider the normalized residuals computed at the solution x$ . We have
postulated that the normalized residuals si are Gaussian random variables with zero mean
and standard deviation 1. Now consider the following variable:
m
= s i2
2
i =1
Since the variables {si, i=1, 2, , m}, are Gaussian random variables, the variable 2 is
also a random variable and it is chi-square distributed [???]. Also since the variables {si,
i=1, 2, , m} are dependent upon only n variables (the state variables x) through a set of
model functions (functions h(x)), the chi-square distributed variable 2 has m-n degrees
of freedom. The chi-square distribution is well known. For example, Table 7.2 tabulates
the probability distribution function of a general chi-square distributed random variable,
Pr( , ) , with degrees of freedom.
Page 27
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
.005
.010
.025
.0000393.000157 .000982
.0100 .0201 .0506
.216
.0717 .115
.484
.207
.297
.831
.412
.554
1.24
.676
.872
1.69
.989
1.24
2.18
1.34
1.65
2.70
1.73
2.09
3.25
2.16
2.56
3.82
2.60
3.05
4.40
3.07
3.57
5.01
3.57
4.11
5.63
4.07
4.66
6.26
4.60
5.23
6.91
5.14
5.81
7.56
5.70
6.41
8.23
6.26
7.01
8.91
6.84
7.63
9.59
7.43
8.26
10.3
8.03
8.90
11.0
8.64
9.54
11.7
9.26
10.2
12.4
9.89
10.9
13.1
10.5
11.5
13.8
11.2
12.2
14.6
11.8
12.9
15.3
12.5
13.6
16.0
13.1
14.3
16.8
13.8
15.0
.050
.100
.250
.500
.750
.900
.950
.975
.990
.995
.00393
.103
.352
.711
1.15
1.64
2.17
2.73
3.33
3.94
4.57
5.23
5.89
6.57
7.26
7.96
8.67
9.39
10.1
10.9
11.6
12.3
13.1
13.8
14.6
15.4
16.2
16.9
17.7
18.5
.0158
.211
.584
1.06
1.61
2.20
2.83
3.49
4.17
4.87
5.58
6.30
7.04
7.79
8.55
9.31
10.1
10.9
11.7
12.4
13.2
14.0
14.8
15.7
16.5
17.3
18.1
18.9
19.8
20.6
.102
.575
1.21
1.92
2.67
3.45
4.25
5.07
5.90
6.74
7.58
8.44
9.30
10.2
11.0
11.9
12.8
13.7
14.6
15.5
16.3
17.2
18.1
19.0
19.9
20.8
21.7
22.7
23.6
24.5
.455
1.39
2.37
3.36
4.35
5.35
6.35
7.34
8.34
9.34
10.3
11.3
12.3
13.3
14.3
15.3
16.3
17.3
18.3
19.3
20.3
21.3
22.3
23.3
24.3
25.3
26.3
27.3
28.3
29.3
1.32
2.77
4.11
5.39
6.63
7.84
9.04
10.2
11.4
12.5
13.7
14.8
16.0
17.1
18.2
19.4
20.5
21.6
22.7
23.8
24.9
26.0
27.1
28.2
29.3
30.4
31.5
32.6
33.7
34.8
2.71
4.61
6.25
7.78
9.24
10.6
12.0
13.4
14.7
16.0
17.3
18.5
19.8
21.1
22.3
23.5
24.8
26.0
27.2
28.4
29.6
30.8
32.0
33.2
34.4
35.6
36.7
37.9
39.1
40.3
3.84
5.99
7.81
9.49
11.1
12.6
14.1
15.5
16.9
18.3
19.7
21.0
22.4
23.7
25.0
26.3
27.6
28.9
30.1
31.4
32.7
33.9
35.2
36.4
37.7
38.9
40.1
41.3
42.6
43.8
5.02
7.38
9.35
11.1
12.8
14.4
16.0
17.5
19.0
20.5
21.9
23.3
24.7
26.1
27.5
28.8
30.2
31.5
32.9
34.2
35.5
36.8
38.1
39.4
40.6
41.9
43.2
44.5
45.7
47.0
6.63
9.21
11.3
13.3
15.1
16.8
18.5
20.1
21.7
23.2
24.7
26.2
27.7
29.1
30.6
32.0
33.4
34.8
36.2
37.6
38.9
40.3
41.6
43.0
44.3
45.6
47.0
48.3
49.6
50.9
7.88
10.6
12.8
14.9
16.7
18.5
20.3
22.0
23.6
25.2
26.8
28.3
29.8
31.3
32.8
34.3
35.7
37.2
38.6
40.0
41.4
42.8
44.2
45.6
46.9
48.3
49.6
51.0
52.3
53.7
[ ]
s
i =1
2
i
( x) = 2 1
where:
m
1 = s i2 ( x )
i =1
Page 28
h ( x ) bi
.
Step 2. Compute the value 1 = s ( x ) = i
i
i =1
i =1
2
Step 3. Read the probability Pr 1 = Pr( 1 , ) from the tabulated chi-square
propability distribution function. Note that interpolation can be used or more detailed
data of the distribution function.
Step 3. Compute the probability Pr 2 1 = 1.0 Pr( 1 , ) .
m
2
i
Note that the confidence level can be computed only for the least squares solution.
However, as an approximation, the above procedure can be applied to the other two
solutions for x$ (L1 and L).
Pr( 1 , ) = Pr 2 1
The accuracy of the solution is expressed with the covariance matrix of the state estimate,
x . Specifically, let x be the true but unknown solution, and x be the solution to the
problem (7.5). This solution may be the least squares, L1 or L solution. The definition
of the covariance matrix is:
C x = E[( x x )( x x ) T ]
Note that a linearized expression for x x is as follows (which is obtained by applying
the state estimation algorithm at point x )
x x = ( H T WH ) 1 H T Wr
Page 29
[ ]
Upon substitution of above into the definition of the covariance matrix, the following is
obtained
C x = E[( H T WH ) 1 H T W T W T H ( H T WH ) T ]
Since the only random variables in above equation is the measurement errors , then
above equation is rewritten as follows:
C x = ( H T WH ) 1 H T W {E T }W T H ( H T WH ) T
Now the above equation is simplified to yield:
C x = ( H T WH ) 1
Once the covariance matrix of the solution has been computed, the standard deviation of
a component of the solution vector x is given with
x = C x (i , i )
i
I = ( H T WH ) 1
Other measures of accuracy can be also derived such as E( b$ ), Cov( b$ ). These variables
are derived next. Consider the estimate of the measurements defined with:
b = h( x )
The statistics of the estimate b$ are computed as follows:
[]
E b = h( x )
Page 30
Note that:
b b = h( x ) h( x ) = H ( x )( x x ) H ( x )( x x )
Upon substitution:
Cov(b) = E H ( x x )( x x ) T H T = H ( H T WH ) 1 H T
Now lets compute the covariance
Cov(b b) = E (b b)(b b) T
Note that:
b b = (b b ) (b b ) = [h( x ) h( x )] r H ( x )( x x ) r = H ( H T WH ) 1 H T Wr r
[]
[ ]
E b = h( x )
E b b = 0
E [J ] = m n
Cov( x ) = C x = ( H T WH ) 1
Cov (b) = H ( H T WH ) 1 H T
Cov (b b) = W H ( H T WH ) 1 H T
Pr 2 1 = 1.0 P( 1 , m n)
In summary, the quality of the state estimate is quantified as follows: A measure of data
validity is expressed with the confidence level obtained from the chi-square distribution.
The accuracy of the estimated state variables is given with the diagonal entries of the
information matrix which express the square of the standard deviation.
The evaluation of the quality of a state estimate is illustrated with an example.
Example E7.5. Consider the problem solved in Example E7.2. Compute the confidence
Page 31
level and the standard deviation of the solution components (V1, V2, 2), as well as the
standard deviation of the measurement estimates given that the error of the measurements
is 0.02, 0.03, 0.03, and 0.04 for the measurements of V1, V2, P2l and Q21, respectively.
Use the least square approach.
Solution: First, the least square technique is used to solve the problem starting with an
initial guess:
x T = [V1 V2
1st Iteration
0.0
0.0
0
r = h( x ) b =
1.56
0.1585
0.0
0.0
1.0
1.0
0.0
0.0
0
H (x ) =
0.0
0.0 14.85
14
.
85
14
.
7
0
.
0
1471.08
H W [h( x ) b] = 1456.22
25740.0
0.0
140327.0 136434.0
H WH = 136434.0 136167.0
0.0
0.0
0.0
245025.0
0.003312
( H WH ) H W [h( x ) b] = 0.007376
0.10505
Page 32
2nd Iteration
0.9967 1.0 0.0033
0
.
08268
0
.
01
0
.
0926764
0.0
0.0
1.0
1.0
0.0
0.0
'
H (x ) =
1.5688 1.5677 14.829
770.865
H W [h( x ) b] = 792.386
139.950
'
0.0017
( H WH ) H W [h( x ) b] = 0.0038
0.000015
'
Page 33
I = ( H WH )
T
V = 2.7583 x 10 -4 = 0.01661
1
V = 2.8435 x 10 -4 = 0.01686
2
= 0.16577 x 10 -4 = 0.00407
2
Note that the standard deviation of the state variables is lower than those of the
measurements.
The confidence level is calculated as follows:
r
1 = i
i =1 i
4
0.0016
0.000475
0.00014
0.0036
=
= 0.020963
+
+
+
0.04
0.03
0.03
0.02
Page 34
the value of the objective function, which can provide information about the quality of
the computed state estimate. For this purpose the chi-square test is utilized.
Measurements
& Telemetry
Noise
Noiseless
Measurements
System
Model
State
Estimator
Measurement Estimate b
p
Covariance of b,
Residual b b
State Estimate x
I
Covariance of x,
Objective Function Value
Probability of Goodness
of Model Fit
(Confidence Level)
Page 35
in the measurement set. Note that the chi-square test does not indicate which datum or
data is bad. The identification of the bad data is achieved with other methods to be
described below.
Page 36
Solution: The equations relating the measurements to the state of the system are linear.
Therefore a linear state estimator provides the solution. The model equations in matrix
form are:
1
2
b = Hx, where : H =
3
1
2
3.1
1
, b=
4
1
1
4.1
0.2891
H T Wb =
2.3602
r = b Hx =
0.7723
0.2843
The sum of the squares of the residuals is:
2
r
J = i = 112.5
1 0 .1
4
The probability of goodness of fit is obtained from Table 7.1 with m-n=2 degrees of
freedom:
Pr 0.0
This probability indicates the presence of bad data in this measurement set. Therefore bad
data have been detected. The next step is to identify the bad data.
The bad data identification will be done with hypothesis testing. Specifically, the
following three hypothesis will be examined:
Hypothesis 1: Measurement 3 (largest residual) is bad.
Hypothesis 2: Measurement 1 (next largest residual) is bad.
Hypothesis 3: Measurement 4 (next largest residual) is bad.
Hypothesis 4: Measurement 2 (next largest residual) is bad.
Page 37
0.289
H T Wb =
2.360
The residuals, sum of weighted squared residuals and probability of goodness of fit with
m-n=1 degrees of freedom (Table 7.1) are:
0.36291
r = b Hx = 0.4355 ,
0.0725
r
J = i = 32.66 ,
1 0 .1
4
Pr 0.0
Note that the probability is still low indicating that measurement 3 may not be a bad
measurement, or there are additional bad measurements in the set.
Hypothesis 2: Measurement 1 is removed and the resulting estimation problem is solved.
The solution is:
x
x = 1 = H TWH
x2
0.15
H T Wb =
3.133
The residuals, sum of weighted squared residuals and probability of goodness of fit with
m-n=1 degrees of freedom (Table 7.1) are:
0.333
r = b Hx = 0.416 ,
0.083
r
J = i = 29.167 ,
1 0. 1
4
Pr 0.0
Note that the probability is still low indicating that measurement 1 may not be a bad
measurement, or there are additional bad measurements in the set.
Hypothesis 3: Measurement 4 is removed and the resulting estimation problem is solved.
The solution is:
x
x = 1 = H T WH
x2
Page 38
1 .0
H T Wb =
1.0333
The residuals, sum of weighted squared residuals and probability of goodness of fit with
m-n=1 degrees of freedom (Table 7.1) are:
0.0333
r = b Hx = 0.0667 ,
0.0333
r
J = i = 0.667 ,
1 0 .1
4
Pr 0.56
Note that the above probability indicates that measurement 4 is a bad measurement and
should be removed from the set of measurements permanently. For this example, there is
no need to proceed further.
Note that the bad measurement was identified in the third hypothesis test. In the original
state estimation solution the residual of the 4th measurement was not the largest
(absolutely). In other words the solution of the initial state estimation problem failed to
yield a high enough residual for this measurement. Only the hypothesis testing was able
to identify the bad measurement. In this case the probability by which the bad datum was
identified is relatively low (0.56) due to the fact that the number of redundant
measurements is very low. One can provide a geometrical interpretation of this data,
observing that the model equations are of the form:
b = ax1 + x2
Graphing the data on a coordinate system where the horizontal axis represents a and the
vertical axis represents b, Figure E7.6 is obtained. In this figure one can clearly see that
three of the data lie on an almost straight line, while the forth datum lies away from this
line (outlier). In this case, the graphical representation of the data can immediately yield
the bad datum.
6
5
4
b 3
2
1
0
0
Page 39
-j15.0
The
0.09940
x =
0.99497
The covariance matrix of the state estimates is:
0.91 0.09
Cov( x ) = 10 6
0.09 0.90
The measurement estimates are:
0.99497
1.48111
b = 0.14914
1.48111
0.00141
The residuals are:
Page 40
.00497
.08111
b b = .01085
.07880
.01141
The value of the objective function at the estimate is:
J ( x ) = 32.688
The low (zero) probability means that bad data exist in the measurement set (detection
step).
Inspection of the residuals reveals that the second or the fourth measurement may be
bad. To determine which one is bad (or maybe both are bad), hypothesis testing is
employed. Three hypotheses will be examined: (1) the fourth measurement is bad, (2)
the second measurement is bad, and (3) both (second and fourth) measurements are bad.
Hypothesis 1. In this case, the fourth measurement is removed. The state estimate
becomes:
.09419
x =
.99496
Above probability is too low. With high probability, this measurement is not a bad
datum.
Hypothesis 2. In this case, the second measurement is removed. The state estimate
becomes:
.10475
x =
.99497
Page 41
0.09940
x =
0.99497
Formulation of the state estimation problem in the min-max sense in terms of the
variables:
V2 = V2 0.99497
2 = 2 + 0.0994
yields:
Minimize r*
Subject to:
V2+ V2 r * + y1 = 0.005
V2+ V2 + r * z1 = 0.005
1.4885V2+ 1.4885V2 14.8509 2+ + 14.8509 2 r * + y2 = 0.0811
1.4885V2+ 1.4885V2 14.8509 2+ + 14.8509 2 + r * z2 = 0.0811
14.926V2+ + 14.926V2 1.4811 2+ + 1.4811 2 r * + y 3 = 0.0009
Page 42
r * 0,
y 0, z 0
V2 = -0.0046
2 = -0.0004
and
V2 = 0.9904
2 = -0.0998
= - 0.000364
= - 0.08
= - 0.0689
= - 0.0798
= 0.08
Using the above computed state and the nonlinear model, the residuals are computed to
be:
rl = - 0.0004
r2 = - 0.0802
r3 = - 0.0679
r4 = - 0.0798
r5 = 0.0787
V2+ V2 + r * z1 = 0.0004
1.4945V2+ 1.4945V2 14.7821 2+ + 14.7821 2 r * + y 2 = 0.0802
1.4945V2+ 1.4945V2 14.7821 2+ + 14.7821 2 + r * z 2 = 0.0802
14.9254V2+ + 14.9254V2 1.4802 2+ + 1.4802 2 r * + y 3 = 0.0679
14.9254V2+ + 14.9254V2 1.4802 2+ + 1.4802 2 + r * z 3 = 0.0679
Page 43
r * 0,
y 0, z 0
V2 = - 0.00009
2 = - 0.0
and
V2 = 0.99031
2 = -0.0998
-r* + y1
-r* + y2
-r* + y3
-r* + y4
-r* + y5
= 0.0003
= 0.08
= 0.0692
= 0.08
= 0.08
Using the above system state and the nonlinear model, the residuals are computed to be:
rl = - 0.00031
r2 = 0.08
r3 = - 0.0693
r4 = - 0.08
r5 = 0.08
The sensitivities of the largest residual with respect to the measurements are obtained
from the linear programming problem (see Appendix B):
dr *
= 0.0 ,
db1
dr *
= 0.5 ,
db2
dr *
= 0.0 ,
db3
dr *
= 0.5 ,
db4
dr *
= 0.0
db5
To determine the quality of the state estimate, the chi-square test is performed. Recall
that the meter accuracy is 2%. Then:
Page 44
0.00031
0.08
0.0693 0.08 0.08
J =
+
+
+
+
= 60.0075
0.02
0.02
0.02 0.02 0.02
This confidence level indicates the presence of bad data. Examining the above data, it is
observed that the second measurement will have the largest effect on the largest residual,
equal to:
dr *
r2 = 0.0401
db2
V2+ V2 + r * z1 = 0.00031
14.9254V2+ + 14.9254V2 1.4800 2+ + 1.4800 2 r * + y 3 = 0.0693
14.9254V2+ + 14.9254V2 1.4800 2+ + 1.4800 2 + r * z 3 = 0.0693
r * 0,
y 0, z 0
Upon solution of the linear program, the system state is computed to be:
Page 45
V2 = - 0.00476
2 = - 0.00459
and
V2 = 0.99507
2 = -0.10439
V2+ V2 + r * z1 = 0.00507
14.9183V2+ + 14.9183V2 1.5553 2+ + 1.5553 2 r * + y3 = 0.0052
14.9183V2+ + 14.9183V2 1.5553 2+ + 1.5553 2 + r * z3 = 0.0052
r * 0,
y 0, z 0
Upon solution of the linear program, the system state is computed to be:
V2 = - 0.000006
2 = - 0.000026
and
V2 = 0.995076
2 = -0.104364
and the residuals as computed with the linear program, the residuals as computed with
the nonlinear model, the sensitivities and the effect of measurements on the largest
Page 46
residual are:
Residual
LP Model
r1
r2
r3
r4
r5
0.005076
--0.005076
0.005076
0.002256
Nonlinear
Model
- 0.005076
--- 0.0051
- 0.0051
0.0023
Sensitivities
dr*/dbi
0.931749
--0.061779
0.006473
0.0
r*
0.004724
--0.000321
0.000030
0.0
To determine the quality of the state estimate, a chi-square test is again performed using
exactly the same performance index. Specifically, rn = 0.10 again.
2
Now m - n = 2.
Reading on the table, the confidence level is:
Pr = 0.952
The jump in the confidence level from zero to 0.952 indicates that the second
measurement is a bad measurement. Note that the min-max approach immediately
identified the bad measurement. Compare this performance with that of the least square
approach, where the bad measurement was identified at the second hypothesis testing.
Page 47
Measurement Set z
Perform State Estimation
Compute Quality of Estimate
Is Estimate
Acceptable?
YES
STOP
NO
Determine Maximum
Normalized Residual sj
Remove Measurement j
Perform State Estimation
Compute Quality of Estimate
Is Performance Improvement
Substantial?
YES
NO
Restore Measurement j
Discard Measurement j
Figure 7.x. Flow Chart for Bad Data Detection, Identification and Rejection
Page 48
measurements on the state estimate with the use of the sequential state estimator. (b) in
bad data identification and especially in hypothesis testing, it is required to evaluate the
effect of removal of a measurement on the state estimate. This problem is ideally suited
to sequential state estimators since with minimal computations one can evalauet the
effects of removing one measurement and therefore can determine if the measurement is
bad or not. The sequential state estimator is presented in this section.
Sequential state estimation algorithms can be developed for general nonlinear systems.
For simplicity we will limit the presentation to linear systems. One can expand the linear
sequential state estimation algorithm to nonlinear systems by linearization of the
problem.
Consider the linear system
r = Hx b
Cov( r ) = W 1
where
b :
x :
r :
The estimate of x in the least square sense, i.e., the one that minimizes the objective
J = r T Wr
is given by
x = (H T WH ) H T Wb
1
Now assume that a new scalar measurement is added to the set of measurements
r ' = hT x b'
E ( r ' ) = 0.0
where:
= A1h
Page 49
r'
+ hT
A = H T WH
a=
Proof: The addition of the new measurement will augment the model as follows:
r H
b
r ' = h T x b'
r W 1
Cov =
r ' 0
H T W
x ' = T
h 0
T
0 H H W
2 h T h T 0
0 b
2 b'
(7.11)
Note that
T
H W
A' = T
h 0
0 H
= H T WH + hh T 2 = A + hh T 2
2 h T
where
A 1
A' 1
A' 1 = A + hh T 2
= A1
1
A1hh T A1
T 1
+h A h
2
Also
T
H W
h T 0
0 b
= H T Wb + 2 b' h
2
b'
1
x ' = A1 2
A1hh T A1 H T Wb + 2 b' h
T 1
+h A h
Page 50
r'
x ' = x 2
T
+h
r'
is a scalar which shall be called a. In this way, equation (7.24) is
+ hT
obtained. The above results suggest the following sequential linear state estimator given
a new measurement of b = hTx:
The quantity
Cov( r ) = W 1
Assume a new measurement ba is added with the following model
ra = ha ( x ) ba
E ( ra ) = 0.0
Page 51
Cov( ra ) = 2
The new state estimate, assuming that one iteration is sufficient, will be:
x ' = x + ( A' ) 1 H T
W
h
0
0 b h( x )
2 ba ha ( x )
where
A :
A =
A :
A=
H :
h :
Application of the matrix inversion lemma as in the case of the linear state estimator, and
noting that
A 1 H T W (b h( x ) ) 0,
yields:
x ' = x a
where
= A 1 h
a=
h' ( x ) b'
2 + hT
The
Example E7.9: Consider the simple electric power system of Figure E7.9 with the
illustrated measuring system. The voltage at bus 1 is known to be 1.0 p.u. without error.
Other measurements taken are
Page 52
~
V1 =V1e j0
-j20.0
V 2 0.97028
.
2839
.
9168
= 0.125197
Page 53
Using Table 7.2, the probability that the data are statistically correct is:
Pr = 0.94
h5 ( x) = Q12 = 0.69111
Thus
b5 h5 ( x) = 0.0089
The linearized model is
hT = [20 V1V2 sin2
- 20 V1 cos2] = [-1.935
-17.9003]
23.110592
= I ( x ) 1 h =
Also
hT
= .00086868
= 0.02
= -.0089/.00086868 = -10.245
.97004
c) The estimates of the measurements are computed at the new state estimate x.
V2 = 0.97004
P12 = 1.9355
P21 = -1.9355
Page 54
Q21 = -0.48446
Q12 = 0.69599
The value of the objective function is computed:
J( x$ ' ) = 0.16645
Using Table 7.2, the probability that the data are statistically correct is :
Pr = 0.9825
The increased confidence is due to the addition of one more good measurement.
is an m-vector of measurements
is an n-vector of states
is an m x n matrix
is an m-vector of measurement noise
A sufficient condition to obtain a unique solution for x in a specified sense (i.e. least
squares sense, LAV sense or min-max sense) is that the rank of H be n. This means that
the sufficient conditions for observability are:
m>n
rank(H) = n
Above conditions are sufficient conditions for algebraic observability. Typically, the
term observability always means algebraic observability.
Page 55
rank ( H ) = rank ( H T H )
The matrix HTH is a square matrix. Now the algebraic observability test becomes
equivalent to determining whether the matrix HTH has full rank. This can be achieved
with an LU factorization procedure as it has been discussed in Appendix A. Specifically,
the matrix H T H is factored into the product of two triangular matrices, L and U.
H T H = LU
Where:
Note that
det{H T H }= det{L}det{U } = l ii u ii ,
i
where:
the symbol means product,
l represents entries of the lower diagonal (L) matrix and
u represent entries of the upper diagonal (U) matrix.
Thus the matrix HTH is full rank if and only if all the diagonal elements of the L and U
matrices are non zero. For further information see Appendix A.
Page 56
Xz
Krumpholz, Clements and Davis [???] have demonstrated with simplified models, that
topological observability is equivalent to algebraic observability for practical power
system estimation problems. Thus, one can use either approach to determine system state
observability. A consequence of this fact is that one can use the topology of the Jacobian
matrix to determine system state observability. As a matter of fact we shall discuss a
method which uses the topology of the information matrix to interpret it as a connectivity
network of the system states. Then using a topology check, the observability can be
determined.
The corresponding topological observability test can be defined as follows. First observe
that the matrix HTH can be interpreted as a network admittance matrix since it is
symmetric. The network has a branch between nodes i and j if and only if the entry ij of
the matrix HTH is non zero. Thus the position of the off-diagonal non zero entries of the
matrix HTH determine the network connectivity of the system states. The system is
observable if and only if this network is fully connected(all states are connected). Highly
efficient algorithms to check connectivity of a network exist, making this approach
practical.
Algebraic and topological observability will be demonstrated with an example.
Example E7.10. Consider the four bus system of Figure E7.10. A number of
measurements are taken as indicated.
3
Voltage Magnitude Measurement
MW Flow Measurement
MVAr Flow Measurement
Page 57
a)
b)
Write the Jacobian matrix H. Only the topology is of interest. Thus set all non zero
entries equal to 1.0.
Write the equation HT(z-b) = (HTH)x and draw a circuit that corresponds to above
equation.
Solution: a)
z = [P12
Q12
P32
Q 32
P24 V1 V 2 V 3 V 4 ]
0
H = 0
1
0
0
0
b)
1 1 0 0 0 0
1 1 0 0 0 0
1 1 1 1 0 0
1 1 1 1 0 0
1 1 0 0 1 1
0 0 0 0 0 0
0 1 0 0 0 0
0 0 0 1 0 0
0 0 0 0 0 1
3
2
T
H H = 0
0
0
0
2 2 0 0 0 0
5 5 2 2 1 1
5 6 2 2 1 1
2 2 2 2 0 0
2 2 2 3 0 0
1 1 0 0 1 1
1 1 0 0 1 2
Page 58
V1
V2
V3
V4
====>
xTHTHx > 0
admittance matrix
nodal voltages
nodal current injections.
Page 59
Since HTH is symmetric, such a network does exist. The construction of this network has
been demonstrated with an example (Example E7.10). When system observability is of
interest the actual values of the entries of the matrix HTH are not important but rather the
topology of the matrix. In this case it suffices to replace the non-zero entries of the
matrix H with arbitrarily selected values such as 1.0.
The identification of the critical measurements can be performed as follows: Consider a
measurement zi. Assume that the matrix HTH has been computed by assigning the value
of 1.0 to the non zero entries of the Jacobian matrix HTH. Observe that the measurement
zi can be expressed as
zi = hiTx
where hiT is the ith row of the Jacobian matrix H. A direct approach to system state
observability will be to remove the measurement zi, recompute the matrix HTH and
perform the observability test. Observe that when the measurement zi is removed, the
Jacobian matrix H is modified to H
0
L
H ' = H hiT
L
0
Upon computation:
HTH = HTH hihiT
The observability test will consist of checking the topology of the matrix HTH. There
may be three cases:
Case 1: The matrices HTH and HTH have the same number of nonzero entries. In this
case, the connectivity of the network of states does not change. Thus, the
measurement zi is not a critical measurement.
Case 2: The matrix HTH has at least two less nonzero entries than matrix HTH. In this
case, the connectivity of the network of states must be examined to determine
observability.
Case 3: The matrix HTH has odd number less nonzero entries than matrix HTH. In this
case, the system is not observable and the connectivity of the network of states
must be examined to determine the unobservable states.
The procedure will be demonstrated with an example.
Page 60
Example E7.11: Consider the system of example E7.10. Determine the type of the
following three measurements: (a) the bus voltage magnitude measurement at bus 4. (b)
The real power flow measurement at circuit 2 - 4. (c) The reactive power flow
measurement at circuit 3 - 2.
Solution: a)
0
0
hh T = 0
0
0
0
0 0 0 0 0 0
0 0 0 0 0 0
0 0 0 0 0 0
0 0 0 0 0 0
0 0 0 0 0 0
0 0 0 0 0 0
0 0 0 0 0 1
Observe that the matrices HTH and HTH have the same number of nonzero entries.
Therefore, the measurement V4 is not a critical measurement.
b) For measurement P24, hT = [0 1 1 0 0 1 1]:
0
0
T
hh = 0
0
0
0
0 0 0 0 0 0
1 1 0 0 1 1
1 1 0 0 1 1
0 0 0 0 0 0
0 0 0 0 0 0
1 1 0 0 1 1
1 1 0 0 1 1
Note that one of the diagonal entries of HTH becomes zero, so P24 is a critical
measurement.
c) For measurement Q32, hT = [0 1 1 1 1 0 0]:
0
0
T
hh = 0
0
0
0
0 0 0 0 0 0
1 1 1 1 0 0
1 1 1 1 0 0
1 1 1 1 0 0
1 1 1 1 0 0
0 0 0 0 0 0
0 0 0 0 0 0
Page 61
Observe that the matrices HTH and HTH have the same number of nonzero entries.
Therefore, the measurement Q32 is not a critical measurement.
This completes the solution.
The above basic structure of the power system state estimation implies the following
assumptions (which in turn result in a biased state estimator):
all current and voltage waveforms are pure sinusoids with constant frequency and
magnitude
the system operates under balanced three phase conditions
the power system is a symmetric three phase system which is fully described by
its positive sequence network
These assumptions introduce deviations between the physical system and the
mathematical model (bias). Mathematically, it is known that the least squares state
estimation procedure is an unbiased estimator if and only if the model is accurate
(exact) and the measuremnent error is statistically distributed. Both of these
Page 62
conditions do not exist in a practical system and have resulted in practical difficulties
manifested by poor numerical reliability of the iterative state estimation algorithm.
Substantial efforts to fine tune the mathematical models in actual field implementations
are required.
To alleviate the sources of error, new measurement systems, power system model and
estimation methods are needed. For example the first assumption can be met by utilizing
synchronized measurements [???]. Synchronization is achieved via a GPS (Global
Positioning System) which provides the synchronizing signal with accuracy of 1 sec.
Assumption 2 can be met by utilizing three phase measurements. Finally assumption 3
can be met by employing full three phase models.
In this section we first discuss the bias resulting from model inaccuracies and then we
discuss the effect of measurment errors. In particular model inaccuracies result from: (a)
unbalanced operating conditions and (b) asymmetries of power system models. Then, a
state estimator is introduced that is based on the following infrastructure:
The state estimation based on this system is not subject to the usual biases of the
traditional state estimation. This state estimation is formulated in its general form that
allows estimation of all three phase voltages resulting in the three-phase state estimation.
z = z t + z
where z t is the true value of the measured quantity (assuming a balanced system), z is
the measurement error due to imbalance, and z is the measurement.
Application of the LS state estimation procedure, assuming no other error sources, yields:
Page 63
x = x t + ( H T WH ) 1 H T Wz
(7.x)
where x t is the true state of the system or the unbiased state estimate, and the second
term is the bias resulting from the imbalance measurement error. Note that the bias from
unbalanced operation depends on the level of imbalance as well as the system parameters
(matrix H).
Page 64
S1 =
1 z max z min
2
z1
S2 =
1 y max y min
2
y1
where z1 is the positive sequence series impedance of the line, zmax and zmin are the max
and min series impedances of the individual phases, y1 is the positive sequence shunt
admittance of the line, ymax and ymin are the max and min shunt admittances of the
individual phases.
The above indices provide in a quantitative manner the level of asymmetry among phases
of a transmission line. As a numerical example, these metrics have been computed for
the line of Figure 7.8 and are presented in Figure 7.9. Note that the asymmetry is in the
order of 5 to 6%.
Page 65
1'-1"
4'
4'
11'-6"
7'-7"
9'-6"
9'-6"
17'-0"
7'-7"
9'-6"
58'-0"
Page 66
Asymmetry Factor
0.06
0.04
Series Admittance
0.02
Shunt Admittance
0.0
180
660
1140
1620
2100
Frequency (Hz)
Because of the presence of non-symmetric components, the state estimate using single
phase measurment set is biased. An estimate of the bias can be computed as follows.
First observe that because of power system component asymmetry, the relationship of a
measurement to the system model will have an error. Specifically:
z = h ( x ) + h ( x )
where h(x) is the function relating the measurement to the state vector assuming
symmetric power system components, h(x) is the difference between the symmetric
model and the asymmetric model.
Now the jacobian matrix of the measurements becomes:
H = H s + H
where H s is the jacobian matrix assuming symmetric power system elements.
Application of the LS state estimation procedure, assuming no other error sources, yields:
x = ( xt + ( H T WH ) 1 H T Wz )( H T WH ) 1 ( I + 2( H T WH )( H T WH ) 1 ) 1 ( H T WH ) (7.x)
where x t is the state of the system assuming a symmetric model, and the other terms
represent the bias resulting from the system asymmetry.
TO BE COMPLETED
Page 67
trends resulted in the use of sensorless technology for power system measurements.
Sensorless technology refers to the use of A/D converter technology to sample the
voltage and current waveforms. Once the sampled waveforms are available, the required
measurements can be retrieved with numerical computations. These systems need
calibration and again the same comments apply for this technology as before.
Independently of the technology used for measurements, it is important to examine
whether there is bias in the measurements. This can be best achieved by examining the
entire measurement channel of a typical power system instrumentation [???]. The major
sources of error (see Figure 7.5a and 7.5b) are (a) the instrument transformers, (b) the
cables connecting the instrument transformers to the sensors or A/D converters and (c)
the sensors or A/D converters. Figure 7.10 illustrates the transfer functions of a typical
instrument transformer. It can be observed that the characteristics of instrument
transformers near the power frequency are flat. One can conclude that for power
frequency measurements, there is no appreciable measurement bias from instrument
transformers. However, cables and A/D converters can introduced appreciable error at 60
Hz. This error will be a systematic error. Figure 7.11 illustrates the transfer function of a
specific A/D converter. Note the magnitude and phase bias even at power frequency. It is
important to note that the measurement bias is dependent upon the design of the A/D
converter. The measurement bias resulting from control cables is variable depending on
the total length of the cables.
Page 68
Figure 7.11. Magnitude and Phase of Frequency Response of the PMU1620 Unit
The measurement bias can be corrected with software. Such methods have been
developed [???], but their use in state estimation is very limited. It is important to note
that the above sources of error cannot be corrected with better (more accurate)
instrumentation. To avoid these sources of error, three phase measurements and a three
phase system model is required.
Page 69
it may provide wrong values, etc. It is very difficult to estimate bounds of the effects of
data time skewness on the state estimator.
The problem can be alleviated by utilization of measurement synchronization. The
technology for this does exist today.
V,I
GPS Receiver
Filter
A/D
V,I
Filter
A/D
V,I
Filter
A/D
PMU
uP
Computer
PMU
Computer
PMU
Computer
Master Station
Monitoring & Analysis
Power System Dynamics
Geomagnetically Induced Currents
Harmonic Generation and Propagation
Page 70
Control
Center
Computer
Use of synchronized measurements simplifies the state estimation problem. Consider for
example the case in which all measurements are GPS-synchronized. We assume here that
the GPS-synchronized measurements have a time precision of better than one
microsecond, the capabil;ity of the GPS technology. Note that in practice this precision
may not be achievable due to other sources of error (see for example reference [***]).
Consider a set of GPS synchronized measurements, ~
z . These measurements are typically
in the form of voltage phasor measurement and electric current phasor measurement. Let
~
x be the vector of the bus voltage phasors. The measured voltage and electric current
phasors will be functions of the voltage phasors, ~
x . In general, these functions will be
linear. We write:
~
z = Y~
x + ~
where Y is a constant matrix and ~ is the measurement error.
The state estimation problem is now formulated as follows:
Min
~i
J =
i i
= ~ H W~
where: ~ = ~
z Y~
x
The above state estimation problem is linear. The solution is obtained by first separating
the equations into real and imagineray parts yielding:
Min J = rT W r + iT W i
Subject to:
r = z r Y r x r + Yi x i
i = z i Y r x i Yi x r
where the subscript r indicates real and the subscript i indicates imaginary.
Upon substitution and observing that at the solution the derivatives of the function J with
respect to the state variables will vanish:
dJ
= 2Y rT W (z r Y r x r + Yi x i ) 2Yi T W (z i Y r x i + Yi x r ) = 0
dx r
dJ
= 2Yi T W (z r Y r x r + Yi x i ) 2Y rT W (z i Y r x i + Yi x r ) = 0
dx i
Solution of above equations for the state vector provides:
Page 71
xr A
x = B
i
B Y rT Wz r + Y i T Wz i
A Y rT Wz i Y i T Wz r
where:
A = Y rT WYr + Yi T WYi
B = Y rT WYi + YiT WYr
The state estimation problem with GPS synchronized measurements will be demonstrated
with an example.
Example E7.x: Consider the two-bus, single transmission line system of Figure E7.x.
The voltage and electric current at the two ends of the line are measured with GPS
synchronized equipment.
0
0
~
~
V1 = 1.010e + j 61.46
pu
V 2 = 0.985e + j 55.09
pu
0
0
~
~
I 1 = 1.456e + j 73.95
pu
I 2 = 1.485e j143.96
pu
The measurement accuracy is 1% for the magnitude and 0.02 degrees for the phase.
Compute the state estimate for this system.
~
V1 =V1e j1
-j15.0
~
V2 = V2e j2
(7.x)
Page 72
~
~
~
V a = V ar + jVai , Vb = Vbr + jVbi , Vc = Vcr + jVci
(7.x)
The set of variables above are state variables, one set for each bus. Here the rectangular
coordinate system is used for convenience.
The number of state variables for a bus are 6n, where n is the number of buses.
z current , a = I ar + jI ai ,
z current ,b = I br + jI bi ,
z voltage, a = V ar + jV ai ,
z current ,c = I cr + jI ci
z voltage,c = Vcr + jVci
Real and reactive power measurements should not be used in the presented three-phase
state estimator for the following reason: since voltage and current is measured, and since
the real power and reactive power is derived from these measurements, all the
information needed is included in the V and I waveform measurements. It certainly does
not mean that the real and reactive power measurements can not be processed.
The measurements, z, are related to the state variables with the equations below.
z current = Yx +
z voltage = Tx +
(7.x)
(7.x)
where all variables are complex variables. Matrix Y is an admittance matrix of proper
dimensions. T is a matrix whose entries are either 1 or 0. If the measured state variables
are ordered first in x in the same order as in z, then matrix T has the form I | 0 with
identity matrix I and zero matrix 0 having proper dimensions. Above equations can be
lumped into the one equation below:
~ ~
~
z = H~
x+r
(7.x)
Note that equation (7.x) is linear. Thus, the least square estimation requires only one
iteration(direct solution). This advantage comes from the use of the rectangular
coordinate system.
Page 73
Minimize: J = ~
r H W~
r
~
~
~
Subject to:
r = z H~
x
(7.x)
Min J = rT W r + iT W i
Subject to:
rr = z r H r x r + H i xi
ri = z i H r xi H i x r
(7.x)
where the subscript r indicates real and the subscript i indicates imaginary.
Upon substitution and observing that at the solution the derivatives of the function J with
respect to the state variables will vanish:
dJ
= 2 H rT W (z r H r x r + H i x i ) 2 H iT W (z i H r x i + H i x r ) = 0
dx r
dJ
= 2 H iT W (z r H r x r + H i x i ) 2 H rT W (z i H r x i + H i x r ) = 0
dx i
Solution of above equations for the state vector provides:
xr A
x = B
i
B H rT Wz r + H iT Wz i
A H rT Wz i H iT Wz r
(7.x)
where:
A = H rT WH r + H iT WH i
B = H rT WH i + H iT WH r
Page 74
~
~ ~
I d = Yd Vd I sd
where
(7.x)
~
I d are the currents at the terminals of the device
~
Vd are the voltages at the terminals of the device
~
I sd are equivalent current sources
Phase Unbiased Measurement means that if there is a measurement for phase A of bus i,
then there are measurements for phases B and C of bus i respectively. The assumption of
Phase Unbiased Measurements simplifies the observability analysis from three phase
case to single phase case, e.g. phase A. This condition is not always true, e.g. in the
presence of neutral current measurements. However, the whole measurement set can be
divided into two parts, (a) the phase unbiased set, and (b) the phase biased set.
Accordingly, the observability analysis can be separated into two steps. Step 1 only
processes phase unbiased measurements.
Page 75
H=
Z 1, A = ~ ,
~x1, A , and ~ = MYA
H
=
x1, A
V1, A
v~1, A
Z% 1,A
x% 1,A
The vector
Z% 1, A ,
~
x1, A
The definition for matrices M, Y, and A can be found in Reference [???]. By proper
v%
ordering of states and measurements: 1,A = In 1 x n 1 0n1 x (n -n1) . In this case,
x% 1,A
H =
where
M
M '=
0
0
,Y ' =
I n1 x n1
Y
0
0
AT
T
,and
A
=
'
I n1 x n1
I n1 x n1 | 0 n1 x (n - n1)
This is equivalent to changing all buses with voltage measurements into a topology with
current flow measurement from the node under consideration to ground. Note that this
procedure permits the use conventional topological methods to perform the step 1 of
observability analysis for harmonic state estimation.
After Step 1, the system is decomposed into three parts as shown in Figure 7.13.
Semi-Observable
Island
Observable
Island
Remaining
Part
Semi-Observable
Island
Page 76
The observable island is an island such that the states at all nodes in this island can be
estimated. The semi-observable island is an island such that if one phase unbiased
voltage measurement is added to any bus in this island, this island can become an
observable island or be merged to the existing observable island.
The remaining part of the system will be unobservable if there is no phase biased
measurement. However, with phase biased measurements, it may be possible to expand
the observable island. This is beyond the scope of this text.
Page 77
case, a hybrid state estimator is emerging. We discuss in this section the hybrid state
estimator.
TO BE CONTINUED
In a real time environment, it is possible that some of the data (measurements) are
inaccurate or bad. In the presence of data redundancy, it is possible to detect (with the
use of the chi-square test), identify (with a combination of consistency checks and
hypothesis testing), and reject bad measurements. This capability is possible whenever
(a) there is enough redundancy of measurements and (b) the redundant measurements are
well distributed throughout the system. Hypothesis testing is computationally demanding
task. The sequential state estimator provides the algorithm for fast hypothesis testing.
In general, the measurements are selected in such a way that the operating state of the
system is observable. However, there are instances where rejection of bad data or local
malfunction of the SCADA system may result in a set of measurements from which the
state is not observable. Techniques for detecting unobservable states have been
presented. In case the system is unobservable, the introduction of pseudo-measurements
provides the means to perform the state estimation and avoid the ill-conditioning of the
problem in this case.
Finally, the assumptions of the traditional state estimation have been reviewed. While
these assumptions are reasonable, nevertheless they introduce biases into the state
estimation problem. These biases have been discussed and their effect has been
quantified. The biases are avoided if three phase estimation methods are used, i.e. the
model of the system is a full three phase model and the measurement set consists of three
phase measurements.
The introduction of synchronized measurements provides the capability to design a noniterative state estimation algorithm.
The state estimator has many other capabilities and chanllenges. The methodologies
canm be used for remote calibration of meters, parameter and topology estimators and
dynamic estimatyion of system oscillations.
Page 78
7.13 Problems
Problem P7.1: Solve the following set of overdetermined equations using least square
estimation:
x 1 + 2 x 2 = 3.05
x1 + x 2 = 2.08
x1 + x 2 = 0.20
2 x 1 + 0.5 x 2 = 2.39
x1 2 x 2 = 0.95
Solution:
x = (H T H ) H T b
1
where:
2
1
3.05
1
2.08
2
H = 1 1 , and b = 0.2
2 0.5
2.39
1 2
0.95
Upon substitution,
0.9669
x =
1.0252
Problem P7.2 Consider the simplified power system of Figure P7.2 with the indicated
measurements. At a specific instance of time, the following readings are taken:
z 1 = V1 = 1.0 pu , 1 = 0.045 pu
z 2 = V 2 = 0.98 pu , 2 = 0.0 pu
Page 79
~
V1 =V1e j0
-j15.0
~
V2 = V2e j2
Figure P7.2
Solution:
0.23
V1
14.7V1 sin 2
,
h(x ) =
14.7V1 sin 2
W = 2 I = 493.82 I
x +1 = x ( H T WH ) 1 H T W (h( x ) b)
Staring from flat start:
Iteration 1:
0.99566
x 1 =
0.11122
Iteration 2:
Page 80
1.00183
x 2 =
0.11125
(b) The information matrix is:
9.2707 0.5136
( H T WH ) 1 = 10 6
0.5136 4.7258
(c)
h ( x ) bi
J ( x ) = 1 = i
i
i =1
4
= 0.2238
Degrees of freedom=4-2=2
From table: Prob.=1.0-0.1053=0.8947
Problem P7.3 Consider the simplified two bus electric power system of Figure P7.3.
The following are measured:
z 1 = V1 = 1.01 pu , 1 = 0.0 pu
z 2 = V 2 = 1.0 pu , 2 = 0.0 pu
Page 81
-j16.0
1.40
0.10
,
b=
1.41
0.10
16.16 sin 2
1
W = diag 2 = diag (1,111.1)
0.06
The second iteration yields:
0.00323
0.12264
= 0.08672
2 = 0.08694 (522.28) 1 [ 16.0989 1.4032 16.0989 1.4032]
0.006766
0.001039
Chi square test:
0.000414 0.12232 0.0104 0.00072
J =
= 16.745
+
+
+
0.03
0.03 0.03 0.03
Pr = 0.0
Page 82
1.40
b = 1.41 ,
0.10
16.16 sin 2
,
h( x ) = 16.16 sin 2
1
W = diag 2
= diag (1,111.1)
2 = 0.08672 (520.32)
0.000414
[ 16.0992 16.0992 1.3995] 0.0104 = 0.08705
0.00072
Page 83
-j16.0
To be continued.
Problem P7.4: Consider an off-nominal tap transformer that belongs to a power system.
The system is illustrated in Figure P7.4. To estimate the value of the tap ratio t, five
measurements are taken as follows:
z 1 = V1 = 0.933 pu , 1 = 0.0 pu
z 2 = V 2 = 0.917 pu , 2 = 0.0 pu
Page 84
Power System
1:t
-j8.0 pu
Figure P7.4
Solution: The problem is formulated with:
z 3 = 6.8445t sin 2
z 4 = 6.8445t sin 2
z 5 = 6.7271 6.8445t cos 2
t
x=
2
+1
x = x ( H T WH ) 1 H T W (h( x ) b)
where:
6.8445 sin 2 6.8445t cos 2
H = 6.8445 sin 2
6.8445t cos 2 ,
6.8445t sin 2
6.8445 cos 2
0
0
1600
W= 0
1600
0
0
1600
0
1.0
x0 =
0.0
1.0 0.0420 0.9580
x1 =
=
0.0 0.0354 0.0354
Problem P7.5: Consider the simplified two bus electric power system of Figure P7.5.
The following are measured:
V1 = 1.01 pu , V 2 = 1.00 pu , P12 = 1.40 pu , P21 = 1.41 pu , Q 21 = 0.10 pu
Compute the weighted least squares estimate of the system state assuming the following
weights: w1 = , w2 = 2500, w3 = 625, w4 = 625, and w5 = 625. Two iterations will
suffice. Note that the first measurement has an infinite weight which means that the error
Page 85
-j16.0
2 ]
Equations:
h1 ( x ) = V 2
h 2 ( x ) = (16.16 )V 2 sin ( 2 )
h3 ( x ) = (16.16)V2 sin( 2 )
x +1 = x ( H T WH ) 1 H T W (h( x ) b)
Two iterations starting from (1,0):
1
V 2
1.00372
=
0.08694
2
2
V 2
0.99940
= 0.08706
Page 86
G1
G2
MW Flow Measurement
MVAR Flow Measurement
kV Measurement
Figure P7.6
Problem P7.7: Consider the following set of overdetermined equations:
x1 + 2 x2 = 3.05
x1 + x2 = 2.08
x1 + x2 = 0.05
2 x1 + 0.5 x2 = 2.39
x1 2 x2 = 0.95
3 x1 x2 = 1.38
Note that the values on the right hand side are measurements. It is known that the
standard deviation of the measurement error is 0.1.
Page 87
(a) Compute the best estimate of the variables x using weighted least square
approach.
(b) Compute the confidence level (chi-square test).
(c) It is suspected that the last measurement is a bad measurement. Perform a
hypothesis testing (that the last measurement is a bad measurement). Use the
sequential state estimator to perform the required calculations. (Hint: the new
information matrix will be: A' = H TWH hhT 2 = A hhT 2 , where A is the
information matrix for the initial problem, A is the information matrix for the
state estimation problem with the last measurement removed, and h is the
coefficients of the last model equation).
Solution:
x = (H T WH ) H T Wb
1
where:
2
1
3.05
1
2.08
1
1 1
0.05
H =
, and b =
2
0
.
5
2.39
1 2
0.95
3 1
1.38
Upon substitution,
0.8903
x =
1.0423
(b) Confidence level:
2
r
J = i = 16.7043, = 6 2 = 4, Pr 0.0
i =1 0.1
6
(c) Hypothesis testing: remove sixth measurement and repeat state estimation.
Method 1: Direct state estimation:
0.9877
1
x = (H 'T W ' H ' ) H 'T W ' b ' =
1.0085
2
5
ri
J =
= 2.4253, = 5 2 = 3, Pr 0.49
i =1 0.1
Page 88
x ' = x +
3
where: h = , 2 = 0.1 2 = 100
1
b6 h T x
hT 2
Computations:
1.6956
0.5875
= A 1 h = 10 3
h T x = 1.6286 b6 h T x = 0.2486
h T = 0.0056743 h T 2 = 0.0043
b6 h T x
0.09744
=
T
2
h
0.03376
0.9877
x ' =
, note same as in part (a).
1.0085
Problem P7.8: Present day technology makes it possible to obtain synchronized
measurements at remote parts of a power system with accuracy of less than one
microsecond using the GPS (Global Positioning System). Synchronized measurements
provide the phasor (magnitude and phase) of the measured quantity. Assume that in a
simplified one line power system synchronized measurements provided the following
measurements
0
0
~
~
V1 = 437.621e + j 61.46 kV
V 2 = 432.508e + j 55.09 kV
0
0
~
~
I 1 = 0.6424e + j 73.95 kA
I 2 = 0.6601e j143.96 kA
~
I1
3.18 Ohms
~
V1
j76.72 Ohms
-j2040 Ohms
~
V2
~
I2
Page 89
The voltage meters have an accuracy of 2% on a 440 kV scale and the electric current
meters have an accuracy of 1% on a 1500 A scale. Formulate the state estimation
problem using Cartesian coordinates, i.e. assume that the state of the system is defined in
T
terms of the real and imaginary part of the bus voltages, i.e. x = [x1 x 2 x 3 x 4 ] ,
~
~
V1 = x1 + jx 2 , and V 2 = x 3 + jx 4 .
a) Compute the best estimate of the state in the least squares sense.
b) Compute the confidence level.
c) Compute the standard deviations of the state variables.
Solution: The admittances are:
1
= 0.00054 j 0.01301 S
g + jb =
3.18 + j76.72
1
jb ' =
= j0.00049 S
j 2040.0
(a) The electric currents as functions of voltages are:
~
~
~
I 1 = (0.00054 j0.01252)V1 (0.00054 j0.01301)V2
~
~
~
I 2 = (0.00054 j0.01301)V1 + (0.00054 j0.01252)V2
The measurements as a function of the state variables are:
~
Re V1 = 209.08 = x1
~
Im V1 = 384.44 = x 2
~
Re V2 = 228.62 = x 3
~
Im V2 = 367.15 = x 4
~
Re I 1 = 0.2444 = 0.00054x1 0.01252x 2 0.00054x 3 + 0.01301x 4
~
Im I 1 = 0.5396 = 0.01252x1 + 0.00054x 2 + 0.01301x 3 0.00054x 4
~
Re I 2 = 0.4738 = 0.00054x1 + 0.01301x 2 + 0.00054x 3 0.01252x 4
~
Im I 2 = 0.3642 = 0.01301x1 0.00054x 2 0.01252x 3 + 0.00054x 4
z = Hx
209.083
384.443
1
T
T
kV
x = (H WH ) H Wb =
247.519
354.679
{
{
{
{
{
{
{
{
}
}
}
}
}
}
}
}
z hiT x
= 0.22014
J = i
i =1 0.02 z b
The confidence level computed from the table with 4 degrees of freedom (8
measurements 4 states=4), is approximately 0.994.
8
Page 90
0
0
1.0 0
0 1.0 0
0
=
0 1.0 0
0
0
0 1.0
0
I = (H T H )
and
V = x2 + x2 = 1.0 ,
1
V = x2 + x2 = 1.0
Problem P7.9: The purpose of this problem is to investigate the impact of phasor
measurements on the state estimation. Consider problem 7.2. Assume that additional
measurements of the voltage phase angle using synchronized measurements were
obtained. The measurement, in effect, adds the following datum 1 2 = 6.4 0 . The
standard deviation of the measurement error is less than 0.080. Note that in this case, the
measurement set consists of synchronized and non-synchronized measurements (hybrid
system).
a) Compute the state estimate;
b) Compute the confidence level;
c) Compute the standard deviation of the state estimate.
Solution: The added measurement is:
0.1115
(b) The confidence level is computed as follows.
h ( x ) bi
J ( x ) = 1 = i
i
i =1
5
= 0.5287
Degrees of freedom=5-2=3
From table: Prob.=1.0-0.09=0.91
(c) The covariance matrix is:
Page 91
0.4107 0.0110
Cov ( x ) = 10 5
0.0110 0.1014
Thus:
V = 0.002, = 0.001
1
Problem P7.10: The purpose of this problem is to investigate the impact of phasor
measurements on state estimation. Specifically, we consider a set of synchronized and
non-synchronized measurements taken on a system and we seek the state estimate by
least squares approach. Consider the simplified two bus system of Figure P7.10. The
following measurements are taken.
Measurement
number
1
2
Measured
Quantity
Measured
Value
Measurement
Error
0.02/0.0004
~
V1 = V1e j1
~
V2 = V2 e j 2
0.98e j1.39 pu
0.02/0.0004
P21
Q21
-1.73 pu
-0.19 pu
0.04
0.04
3
4
1.0e j1.51
pu
Note that the first two measurements are phasor measurements (complex), i.e. they
are synchronized measurements.
a)
b)
c)
d)
Formulate the state estimation problem for above system. Hint: Each synchronized
measurement can be considered as two measurements: measurement of the rms value
and of the phase angle. Also define the state of the system as x1 = V1 , x 2 = V 2 and
x3 = 1 2 .
Compute the state estimate.
Compute the confidence level.
Compute the standard deviation for all states.
~
V1 =V1e j0
-j15.0
~
V2 = V2e j2
Figure P7.10
Page 92
z 3 = 0.12 = + 3
+
+
0.02
0.04
0.0004
i =1
i =4
2
Min
x +1 = x ( H T WH ) 1 H T W (h( x ) b)
1
H =
0
15V 2 sin
15V 2 cos
W = diag (2500
0
1
0
15V1 sin
30V 2 15V1 cos
2500 6250000
15V1V 2 cos
15V1V 2 sin
0
625 625 )
Start from:
1.0
0
x = 0.98
0.12
2
h ( x) zi
= 0.0099 . From tables Prob=0.995 (degrees of freedom: 5-3=2)
(c) J = i
i
i =1
I = (H T WH )
0.1925 0.0445
0.1898
0.2030 0.0470
= 10 3 0.1925
Thus:
V1 = 0.0138, V2 = 0.0142, = 0.00043
Page 93
~
V1
0
~
V2 = 129.6e j1.98 kV
0
~
I 2 = 0.558e j179.86 kA
~
I1
jbS
jb
~
I2
~
V2
jbS
The voltage meters have an accuracy of 2% on a 440 kV scale, the electric current meters
have an accuracy of 1% on a 1500 A scale and the time precision is better than 1
microsecond which translates as a precision of phase angle of 0.02 degrees. Formulate
the state estimation problem using Cartesian coordinates, i.e. assume that the state of the
system is defined in terms of the real and imaginary part of the bus voltages, i.e.
~
~
T
x = [x1 x 2 x 3 x 4 ] , V1 = x1 + jx 2 , and V 2 = x 3 + jx 4 .
a) Compute the best estimate of the state in the least squares sense.
b) Compute the confidence level.
c) Compute the standard deviations of the state variables.
d) The above approach does not recognize the preciosion of the phase measurements. To
account for this precision, reformulate the state estimation problem in polar coordinates,
T
i.e. the state variables will be: x = [V1 V2 1 2 ] (only three state variables). Repeat
(a), (b) and (c) for this formulation.
e) State your observations.
Example E7.7. Consider the system of Figure E7.7. Five measurements are taken as
follows: V2 = 0.99, Pl2 = 1.4, Ql2 = 0.15, P2l = -1.46, Q21 = 0.01.
Page 94
-j15.0
All measurement instruments are known to have an accuracy of 1%. The voltage at bus 1
is 1.0 p.u. with absolute certainty. Perform bad data identification.
Page 95