Dr R Jegatheesan

Professor, EEE Dept.

SRM University

PS2106

POWER SYSTEM STATE ESTIMATION


UNIT I - INTRODUCTION TO STATE ESTIMATION (9 hours)

Need for state estimation - Comparison of power flow and state estimation
problems - Measurements - Redundancy - Noise - Measurement functions -
Weighted least square errors - Measurement Jacobian matrix - Weights - Gain
matrix - State estimation applied to DC networks - Bad data deduction and
removal.


UNIT II - POWER SYSTEM STATE ESTIMATION (9 hours)

Building model for power system components - Mean - Standard deviation -
Maximum likelihood estimation - Measurement equations - WLS State Estimation
- Measurement Jacobian matrix - Gain matrix - Cholesky decomposition -
Forward and backward substitutions - State estimation algorithm - Decoupled
formulation of WLS state estimation - DC State estimation.
UNIT III - NETWORK OBSERVABILITY ANALYSIS (9 hours)

Networks and graphs - Methods of observability analysis - Numerical method
based on nodal variables - Determining the unobservable branches -
Identification of observable islands - Measurement placement to restore
observability.

UNIT IV - BAD DATA DETECTION AND IDENTIFICATION (9 hours)

Properties of measurement residuals - Classification of measurements - Bad data
deduction using Chi-square distribution - use of normalized residuals for bad
data deduction - Bad data identification by largest normalized residual test.
Hypothesis testing identification.
UNIT V - PHASOR MEASUREMENT UNITS (9 hours)

Basics of Phasor Measurement Unit (PMU) - Optimal placement of PMUs -
Methods to reduce the number of PMUs. State estimation including PMU -
Optimal placement of PMUs in large scale systems - State estimation including
FACTS devices.
REFERENCES

1. Ali Abur and Antonio Gomez Exposito, Power System State Estimation
Theory and Implementation, Marcel Dekker. Inc., New York, Basel 2004.

2. J J Grainger and W D Stevension, Power System Analysis, McGraw-Hill,
Inc., 1994.

3. A Monticelli, State Estimation in Electric Power Systems, Kluwer
Academic Publishers, 1999.

4. Mukhtar Ahmad, Power System State Estimation, LAP Lambert Acad
Publishers, 2013.

5. Naim Logic. Power System State Estimation, LAP Lambert Acad
Publishers, 2010.
Chapter 1

INTRODUCTION TO STATE ESTIMATION

Power systems are composed of generation, transmission, sub-transmission and
distribution systems. Power is injected into the system by the generators and
consumed by the loads. In between the generation system and loads there exists
large complex transmission and distribution system.
The output voltages of the generators are in the range of 11 to 25 kV.
Transformers are used to increase the voltage levels to levels ranging from 220
kV all the way up to 765 kV at the generating stations for efficient power
transmission. High voltage is preferred at the transmission system for different
reasons one of which is to minimize the copper losses that are proportional to the
current flows in the lines.
At the receiving end, the transmission systems are connected to the sub-
transmission and distribution systems which are operated at lower voltage levels
ranging from 132 kV to 0.44 kV. Distribution systems are typically configured to
operate in a radial configuration, where the feeders stretch from distribution
substations and form a tree structure with their roots at the substation and
branches spreading over the distribution area.
1.1 OPERATING STATES OF A POWER SYSTEM
Knowing the network model and the complex phasor voltages at every
system bus the operating conditions of a power system at a given
point in time can be determined. Since the set of complex phasor
voltages fully specifies the system, it is referred as the static state of
the system.
As operating conditions change, the system may move into one of the
five possible states namely, normal, alert, emergency, in extremis and
restorative. Figure 1.1 shows the different operating states of a power
system and the transitions between them.















Restorative

Alert

In extremis
Emergency

Normal
Fig. 1.1 Power system operating states
Fault clearing
Excitation control
Fast valving
Load shedding

Load shedding
System separation

Generation shifting
Increase reserve

Emergency
NORMAL STATE:
In the normal state, all system variables are within the normal range and no
component is being overloaded. The system operates in a secure manner and is
able to withstand a contingency without violating any of the constraints.
ALERT STATE:
The system enters the alert state if the security level falls below a certain limit of
adequacy, or if the possibility of disturbance increases because of adverse
weather conditions such as the approach of severe storms. In this state, all the
system variables are still within the acceptable range and all the constraints are
satisfied. However, the system has been weakened to a level where a contingency
may cause an overloading of a component that places the system in an
emergency state. If the disturbance is very severe, the in extremis state may result
directly from the alert state.
RESTORATIVE STATE:
Preventive action, such as generation shifting or increased reserve, can be taken
to restore the system to the normal state. If the restorative steps do not succeed,
the system will remain in the alert state.
EMERGENCY STATE:
The system enters the emergency state, if a sufficiently severe disturbance
occurs when the system is in the alert state. In this state, voltages at many buses
are low and / or component loadings exceed the short-term emergency ratings.
The system is still intact and may be restored to the alert state by initiating
emergency control actions such as fault clearing, excitation control, fast-valving
and load curtailment. If the above measures are not applied or are ineffective, the
system will move to in extremis state.
IN EXTREMIS STATE:
If the control action taken during the emergency state is insufficient, then the
system enters into in extremis state. The result is cascading outages and
possibly a shut-down of major portion of the system. Control actions, such as
load shedding and controlled system separation, are aimed at saving as much
system as possible from a widespread blackout.
In case the control action taken are effective, the system moves to restorative
state in which further action is being taken to reconnect all the facilities and to
restore the system load. The system transits from the restorative state to either
alert state or normal state, depending on the system conditions.
1.2 POWER SYSTEM SECURITY ANALYSIS
Power systems are operated by system operators from the area control centers.
The main goal of the system operator is to maintain the system in a normal
secure state as the operating conditions vary during the daily operation.
Following are required to achieve this goal requires
i) Continuous monitoring of the system conditions
ii) Identification of the operating state and
iii) Determination of the necessary preventive action in the case of state found
to be insecure.
This sequence of operation is referred as the security analysis of the system.
The first step of security analysis is to determine the current state of the system.
This involves acquisition of measurements from all parts of the system.
The measurement may be both of analog and digital type.
Substations are equipped with devices called Remote Terminal Unit (RTU) which
collect various types of measurements from the field and transmit them to the
control center.
More recently, the so-called Intelligent Electronic Device (IED) are replacing or
complementing the existing RTUs.
Once the data are collected at the they are processed at the control center in
order to determine the system state.
It is possible to have a mixture of these devices (RTUs and IEDs)
connected to a Local Area Network (LAN) along with Supervisory
Control And Data Acquisition (SCADA) front end computer, which
supports the communication of the collected measurements to the
host computer at the control center.
The SCADA host computer at the control center receives
measurements from all SCADA systems installed at the monitored
substations via one of many possible types of communication links
such as fiber optic, satellite, microwave, etc.
Fig.1.2 shows the configuration of EMS / SCADA system for a typical
power system.
CONTROL
CENTER
Local Area Network
Monitored Devices Substation
Fig. 1.2 EMS / SCADA system configuration
Communications
Network
CORPORATE
OFFICES
Fig.1.2 shows the configuration of EMS / SCADA system for a typical power
system.










Planning
and
analysis
functions
SCADA host computer
Energy Management System
(EMS) Application Functions
State Estimator
SCADA Front end computer
RTU RTU IED RTU IED
Measurements received at the control center will include line power flows, bus
powers, bus voltage magnitudes, line current magnitudes, generator output
powers, loads powers, circuit breaker and switch status information, transformer
tap positions and switchable capacitor bank values.
These raw data and measurements are processed by State Estimator (SE) in
order to filter the measurement noise and also detect gross errors if any. Thus,
State estimator solution will provide reliable estimate of the system state based
on the available measurements and on the assumed system model.
The information pertaining to the system state will then be passed on to all the
Energy Management System (EMS) application functions such as the
contingency analysis, automatic generation control, automatic load frequency
control, economic load dispatching, load forecasting and optimal power flow, etc.
The same information will also be available via a LAN connection to the corporate
offices where other planning and analysis functions can be executed off-line.
Development of EMS that encompass State Estimation
Initially, power systems were monitored only by supervisory control systems
which essentially monitor and control the status of circuit breakers at the
substations. Generator outputs and the system frequency were also monitored
for purpose of Automatic Generation Control (AGC) and Economic Dispatch (ED).
These supervisory control systems were later augmented by real time data
acquisition capabilities. This allows the control centers to gather all sorts of
analog measurements and circuit breaker status data from the power system.
This led to the establishment of Supervisory Control and Data Acquisition
(SCADA) systems.
The main motivation behind the development of SCADA system was the
facilitation of security analysis. Various operating functions can be executed only
by knowing the real-time operating conditions of the system.
However, the information provided by the SCADA system may not always be
reliable due to the errors in the measurements, telemetry failures, communication
noise, etc.
Furthermore, the collected set of measurements may not allow direct extraction
of the corresponding A.C. operating state of the system. For instance, bus voltage
phase angles are not typically measured and not all the transmission line flows
are available.
Besides, it may not be economically feasible to telemeter all possible
measurements even if they are available.
1.3 STATE ESTIMATION
The idea of state estimation in power systems was first recognized in the year
1970 and subsequently addressed by Fred Schweppe.
Introduction of state estimation function broadened the capabilities of the SCADA
system computers, leading to the establishment of Energy Management System
(EMS).
State estimators provide a reliable real-time data base of the system, including the
existing state, based on which necessary security assessment functions can be
deployed in order to analyze contingencies, and to determine any required
corrective actions.
The state estimators typically include the following functions
Topology processor: Gathers status data about the circuit breakers and switches
and configures the one-line diagram of the system.
Observability analysis: Determines if a state estimation solution for the entire
system can be obtained using the available set of measurements. Identifies the
unobservable branches and the observable islands in the system if any exist.
State estimation solution: Determines the optimal estimate for the system state,
which is composed of complex bus voltages in the entire power system, based
on the network model and the gathered measurements from the system.
Bad data processing: Detect the existence of gross errors in the measurement
set. Identifies and eliminates bad measurements provided that there is enough
redundancy in the measurement configuration.
Parameter and structural error processing: Estimates various network
parameters such as transmission line model parameters, tap changing
transformer parameters, shunt capacitor or reactor parameters. Detects
structural errors in the network configuration and identifies the erroneous
breaker status.
APPLICATION FUNCTIONS
RAW
MEASUREMENTS

RELIABLE
STATE
Thus Power System State Estimator constitutes the core of
the on-line security analysis function.
It acts as a filter between the raw measurements received
from the system and all the application functions that require
the most reliable data base pertaining to the current state of
the system.



STATE ESTIMATOR
APPLICATION
FUCTIONS
DIFFERENCE BETWEEN POWER FLOW AND
STATE ESTIMATION
Power system networks are large and complex.

For proper operation of power system, several
control actions are necessary.

To decide about the control actions, system
status need to be determined first.

System state can be decided by conducting
Power flow analysis.
POWER FLOW ANALYSIS
If even one of the inputs is unavailable, power flow
solution can not be obtained.
Gross errors in one or more input quantities can cause
power flow results to become useless.
POWER SYSTEM NETWORK
P and Q injections
Bus voltages

V

and
STATE ESTIMATION
POWER SYSTEM NETWORK
Bus voltages

V

P and Q injections
Bus voltages
P and Q line flows

V
Conveniently
measurable
and
Number of inputs > Number of state variables.
State variables are estimated.
Bad data if any can be detected and removed.
Let us say that we like to determine the area of a rectangular field.
Area A = L x B
We can measure L and B and compute the area.
But both the measurements may have certain errors and hence area calculated is
not accurate.
To overcome this we can measure
1. Length L
2. Breadth B
3. Diagonal D
4. Perimeter P
From these four measurements area A can be estimated filtering out the errors
and the area obtained in this way will be more accurate and reliable.
The following are the features of State Estimation
1. The techniques developed provide an estimate of the
system state.
2. Further they provide a quantitative measure of how good
the estimate is before it is used for real-time power
flow calculations.
3. The inputs are easily measurable P and Q line flows, P
and Q bus power injections and voltage magnitudes V.
4. The unavoidable errors of the measurements are
assigned statistical properties and the estimates of the
states are subjected to statistical testing before being
accepted as satisfactory.
THE METHOD OF WEIGHTED LEAST SQUARES
In the power system, meters are used to measure real power,
reactive power, voltages and currents.
The analog quantities measured at substations and other
strategical points, pass through transducers and analog-to-
digital converters and the digital outputs are then
telemetered to energy control center.
The data received at energy center are processed by the
computer.
The acquired data always contain inaccuracies.
The errors can be quantified in a statistical sense and the
estimated values are then either accepted as reasonable or
rejected based on the accuracy.
Because of noise, the true values of physical quantities are
never known.
The method of least squares is often used to best fit
measured data relating two or more quantities.
First, we apply the method to a simple dc circuit which
contains measurement errors.
Later we extend the estimation procedures to the ac power
system.
The best estimates are chosen as those which minimize the
weighted sum of squares of the measurement errors.
This method is known as Weighted Least Square ( WLS )
method.
Consider the simple dc circuit of Fig. 1 below.
1
z
Here and are the unknown values to be
estimated.
They are represented by the state variables
, , and are the four measurements made.
In terms of network parameters and the state
variables, true values of measurements can be written
as (using Super position theorem)
1
V
2
V
1
z
.
2
x and
1
x
V
A
V
A
1
1 1
1 1
-
+
+
+ +
+
+
-
- -
-
- 1
V 2
V
4
z
3
z
2
z
3
z
2
z
4
z
Each measurement will contain some error . Thus measurements
are given by
(1)
(2)
2 1 true 4
2 1 true 3
2 1 true 2
2 1 true 1
x
8
3
x
8
1
z
x
8
1
x
8
3
z
x
8
5
x
8
1
z
x
8
1
x
8
5
z
+ =
+ =
+ =
=
4 2 1 4
3 2 1 3
2 2 1 2
1 2 1 1
e x
8
3
x
8
1
z
e x
8
1
x
8
3
z
e x
8
5
x
8
1
z
e x
8
1
x
8
5
z
+ + =
+ + =
+ + =
+ =
i
z
i
e
The above equation can be written as
(3)
i.e.
(4)
In a compact form z = H x + e
Error e = z -
= z - H x
(5)
(6)
True values of can not be determined.
However, we can calculate the estimates of .
true
z
2 1
x and x
. .
2 1
x and x
4 true 4 4 2 42 1 41 4
3 true 3 3 2 32 1 31 3
2 true 2 2 2 22 1 21 2
1 true 1 1 2 12 1 11 1
e z e x h x h z
e z e x h x h z
e z e x h x h z
e z e x h x h z
+ = + + =
+ = + + =
+ = + + =
+ = + + =
(
(
(
(

+
(

(
(
(
(

=
(
(
(
(

4
3
2
1
2
1
42 41
32 31
22 21
12 11
4
3
2
1
e
e
e
e
x
x
h h
h h
h h
h h
z
z
z
z
e = z - H x (6)

Just as
true
z = H x we can write,

.
z = H
.
x
where
.
x is the estimated state values and

.
z is the estimated measurements.

Estimated errors are obtained as
. .
= z - z e
.
= x H - z (7)

Substituting z = H x + e in eq. (7)
=
.
e H x + e - H
.
x = e - H (
.
x - x ) (8)
PROCEDURE
First find the estimated state
.
x
From this compute estimated measurement
.
z using
.
z = H
.
x
Finally find estimated error
.
e using
. .
= z - z e
.
= x H - z
It is to be noted that
.
x ,
.
z and
.
e are related.
We need to find the state variables for which weighted sum of squares
of errors is minimum. i.e. it is required to minimize
2
4 4
2
3 3
2
2 2
2
1 1
2
j
4
1 j
j
e w e w e w e w e w f + + + = =

=
(9)
Noting that error is related to and necessary conditions for f
to be minimum are
.
x = 0
]
x
e
e w
x
e
e w
x
e
e w
x
e
e w [ 2
x
f
21
4
4 4
2
3
3 3
2
2
2 2
2
1
1 1
2
c
c
+
c
c
+
c
c
+
c
c
=
c
c
]
x
e
e w
x
e
e w
x
e
e w
x
e
e w [ 2
x
f
1
4
4 4
1
3
3 3
1
2
2 2
1
1
1 1
1
c
c
+
c
c
+
c
c
+
c
c
=
c
c
.
x
.
x
.
x = 0
(10)
(11)
(

=
(
(
(
(
(
(

(
(
(
(

(
(
(
(

c
c
c
c
c
c
c
c
c
c
c
c
c
c
c
c
.
.
.
.
0
0
e
e
e
e
w
w
w
w
x
e
x
e
x
e
x
e
x
e
x
e
x
e
x
e
4
3
2
1
4
3
2
1
2
4
2
3
2
2
2
1
1
4
1
3
1
2
1
1
(12)
.
x
i
e
1
x
2
x




(
(
(
(
(
(
(
(
(

c
c
c
c
c
c
c
c
c
c
c
c
c
c
c
c
2
4
1
4
2
3
1
3
2
2
1
2
2
1
1
1
x
e
x
e
x
e
x
e
x
e
x
e
x
e
x
e
= -
(
(
(
(

42 41
32 31
22 21
12 11
h h
h h
h h
h h
= - H where matrix H =
(
(
(
(

42 41
32 31
22 21
12 11
h h
h h
h h
h h

Note that the elements of H matrix are independent of state variables x
1
and x
2
. This is true only
when the measurements are LINEARLY related to state variables as shown in eq. (1)

2 42 1 41 4 4
2 32 1 31 3 3
2 22 1 21 2 2
2 12 1 11 1 1
x h x h z e
x h x h z e
x h x h z e
x h x h z e
(3) eq. From
=
=
=
=
T
Using the above result in eq.(12) we have
(
(
(
(

42 41
32 31
22 21
12 11
h h
h h
h h
h h

(

=
(
(
(
(
(
(

(
(
(
(

.
.
.
.
0
0
e
e
e
e
w
w
w
w
4
3
2
1
4
3
2
1
(13)
We are more interested on
.
x rather than
.
e. We can replace
.
e by
.
x .
Using compact notation
MATRIX GAIN the as called is G
H W H G where
(14) z W H G x Therefore
z W H x H W H Thus
0 ) x H z ( W H i.e. 0 ) z z ( W H i.e. 0 e W H
T
T 1
T T
T T T
=
=
=
= = =

.
.
. . .

We expect
.
X to be close to the true values of X. Let us now develop an expression for
.
X - X.
Substituting z = H X + e in eq.(14) namely z W H G x
T 1
.
=
(15) e W H G X X Thus
e W H G X e W H G X H W H G ) e X H ( W H G X
T 1
T 1 T 1 T 1 T 1

.

.
=
+ = + = + =

For the example circuit
(
(
(
(

=
(
(

.
.
4
3
2
1
2 2
1 1
e
e
e
e
* * * *
* * * *
x x
x x
(16)
This means that any one or more of the four errors
4 3 2 1
e and e , e , e can influence the
difference between each of the state estimate and its true value. i.e. WLS calculation
spreads the effect of the error in any one measurement, to some or all the estimates.
Similarly, we can compute the estimated error as (using Eq. 15)

(17) e ) W H G H I (
e ) W H G ( H e ) X X ( H e X H e X H z z e
T 1
T 1

. . . .
=
= = + = =


Use of eqs. (15) and (17) will be given after introducing the statistical

properties of errors.

Summary of formulas:

z = H X + e

e = z - H X

Using WLS method, z W H G X
T 1
.
= where G = H W H
T


.
z = H
.
x

) X X ( H e X H e X H z z e = + = =
. . . .

Procedure for DC State Estimation:
1. Generally a set of measurements and weighting factors of errors will be the
input data.
2. Relate each measurement to the state variables and thereby find the H
matrix.
3. Compute the gain matrix G from G = H
T
W H
4. Determine estimated state variables from z W H G x
T 1
.
=
5. Compute
. .
= x H z

6. Calculate the estimated errors from
. .
= z - z e
Example 1

In the dc circuit of Fig.1, the meter readings are
1
z = 9.01 A,
2
z = 3.02 A,
3
z = 6.98 V and
4
z = 5.01 V. Assuming that the ammeters are more
accurate than the voltmeters, let us assign the measurement weight
100 w
1
= , 50 w and 50 w 100, w
4 3 2
= = = respectively. Determine WLS
estimates of the voltage sources
2 1
V and V . Also calculate the estimated
measurements and the estimated errors in the measurements.

Solution

Given ;
50
50
100
100
W ;
5.01
6.98
3.02
9.01
z
(
(
(
(

=
(
(
(
(

=

it is required to find
. . .
e and z , x .
For the DC circuit shown in Fig. 1, measurements are related to the state
variables as





Thus
(
(
(
(

=
0.375 0.125
0.125 0.375
0.625 0.125
0.125 0.625
H
4 2 1 4
3 2 1 3
2 2 1 2
1 2 1 1
e x
8
3
x
8
1
z
e x
8
1
x
8
3
z
e x
8
5
x
8
1
z
e x
8
1
x
8
5
z
+ + =
+ + =
+ + =
+ =
Gain matrix G = H W H
T

W H
T
=
(

0.375 0.125 0.625 0.125
0.125 0.375 0.125 0.625
(
(
(
(

50
50
100
100

=
(

18.75 6.25 62.50 12.50
6.25 18.75 12.50 62.50


Gain matrix G =
(

18.75 6.25 62.50 12.50
6.25 18.75 12.50 62.50

(
(
(
(

0.375 0.125
0.125 0.375
0.625 0.125
0.125 0.625

=
(

48.4375 10.9375
10.9375 48.4375


Its inverse
1
G

=
(

0.0218 0.0049
0.0049 0.0218

Estimated state vector

z W H G X
T 1
.
=
=
(

0.0218 0.0049
0.0049 0.0218

(

18.75 6.25 62.50 12.50
6.25 18.75 12.50 62.50

(
(
(
(

5.01
6.98
3.02
9.01

=
(

0.4386 0.2281 1.2982 0.0351

0.2281 0.4386 0.0351 1.2982

(
(
(
(

5.01
6.98
3.02
9.01


=
(

V 8.0261
V 16.0072

G
-1
H
T
W
Estimated measurements

. .
= X H Z =
(
(
(
(

0.375 0.125
0.125 0.375
0.625 0.125
0.125 0.625

(

8.0261
16.0072
=
(
(
(
(

5.01070
7.00596
3.01544
9.00123



Estimated errors

. .
= z z e =
(
(
(
(

5.01
6.98
3.02
9.01
-
(
(
(
(

5.01070
7.00596
3.01544
9.00123
=
(
(
(
(

V 0.00070 -
V 0.02596 -
A 0.00456
A 0.0877

In the proceeding example the state of the system is estimated using WLS

method. The results are of little values if we cannot measure, by some

means, how good the estimate is.

Consider for instance, that the voltmeter reading
4
z of Fig.1 is 4.40 V

rather than 5.01 V used in Example 1. If other three meter readings are

unchanged, we can determine
.
X ,
.
z and
.
e as follows.

z W H G X
T 1
.
=
=
(

0.4386 0.2281 1.2982 0.0351

0.2281 0.4386 0.0351 1.2982

(
(
(
(

4.40
6.98
3.02
9.01

=
(

V 7.75860
V 15.86807

. .
= X H z =
(
(
(
(

0.375 0.125
0.125 0.375
0.625 0.125
0.125 0.625

(

7.75860
15.86807
=
(
(
(
(

4.89298
6.92035
2.86562
8.94772


. .
= z z e =
(
(
(
(

4.40
6.98
3.02
9.01
-
(
(
(
(

4.89298
6.92035
2.86562
8.94772
=
(
(
(
(

V 0.49298 -
V 0.05965
A 0.15438
A 0.06228


It is meaningless to compare the results obtained in the above two cases,
because at any one time, only one set of measurement will be available.

Given a set of measurements, how we can decide if the corresponding
results should be accepted as good estimates of the true values? What
criterion for acceptance is reasonable?

If a grossly erroneous meter reading is present, can we detect that fact and
identify the BAD DATA ?
STATISTICS, ERRORS AND ESTIMATES
Generally unavoidable random noise enters into measurement
process and distorts the results from the true values.
Repeated measurements of the same quantity reveals certain
statistical properties from which the true value can be estimated.
If the measured values of a random variable are plotted, taking
the measured values along the horizontal axis and their relative
frequency of occurrence along the vertical axis, a bell shaped
function is obtained. This function is called the Gaussian or
normal probability density function and has the formula
(18)
2
)

x
(
2
1

2
1
f(x)

=
where x: randam variable
: mean or expected value of x = E(x)
: standard deviation of x
This function will have the maximum value when x =


The probability of x to lie between to + is 0.342
Area under the curve gives the probability with the corresponding
intervals of the horizontal axis.

f(x) has its maximum value when x equals , which is the expected
value of x, denoted by E(x) and defined by
}

= = dx ) (x f x ) (x E
The expected value is often called the MEAN.
(19)
The degree to which the curve f(x) spreads out about depends on
the variance of x defined by
2

dx ) x ( f ) x ( ] ) x ( [ E
2 2 2
}

= = (20)
The positive square root of the variance is the standard deviation
of x.
The total area of the curve is one.
We assume that the noise terms
4 3 2 1
e and e , e , e are independent Gaussian
random variables with zero means and the respective variances are
. and , ,
2
4
2
3
2
2
2
1


Two random variables
j i
e and e are independent when E (
j i
e e ) = 0 for j i = .

Later we encounter the product vector, e and its transpose
T
e . This is
given by

e
T
e =
(
(
(
(

4
3
2
1
e
e
e
e
| |
(
(
(
(
(

=
2
4 3 4 2 4 1 4
4 3
2
3 2 3 1 3
4 2 3 2
2
2 1 2
4 1 3 1 2 1
2
1
4 3 2 1
e e e e e e e
e e e e e e e
e e e e e e e
e e e e e e e
e e e e (21)


The expected values of diagonal elements are nonzero and correspond to
the variance. Then the expected value of Eq.(21) becomes

E ( e
T
e ) = R =
(
(
(
(
(

2
4
2
3
2
2
2
1
0 0 0
0 0 0
0 0 0
0 0 0
(22)

We know
1 true 1 1
e z z + =

The shift of e
1
by z
1 true
in no way alter the shape and spread of the
probability density function of e
1
. Hence, z
1
also has a Gaussian
probability density function with a mean value
1
equal to the true value
z
1 true
and variance
1
2
equal to that of e
1
. Similar remarks are applicable to
other three measurements also.

Meters with smaller error variances have narrower curves and provide
more accurate measurements. In formulating the objective function f of
Eq. (9), preferential weighting is given to more accurate measurements by
choosing the weight w
i
as the reciprocal of the corresponding variance
i
2
.
Henceforth, we specify the weighting matrix W of Eqs. (15) and (17) as

(
(
(
(
(
(
(
(
(

= =

2
4
2
3
2
2
2
1
1

1
R W (23)

and the gain matrix G then becomes

H R H G
1 T
= (24)

In Example 1, the weight for the first ammeter is taken as 100. This means
that its variance is 1/100 = 0.01. Thus its standard deviation is 0.1. This
means that there is 99% probability for this ammeter, when functioning
properly, to give readings within 0.3 A (3 = 0.3 A) deviation from the true
values of its measured current. Similar interpretation can be given for the
other meters.
Let us see how close the estimated value is to the true value.
From Eq. (19),
which defines the expected value the random variable x, we can show that
E[ax + b] = a E[x] + b if a and b are constants. Therefore, taking the
expected value of both side of Eq.(15) namely
(15) e W H G X X
T 1
.
= gives

(
(

] x x [ E
] x x [ E
2
^
2
1
^
1
=
(
(

2
^
2
1
^
1
x x [ E
x x [ E
]
]
= G
-1
H
T
R
-1

(
(
(
(

] e [ E
] e [ E
] e [ E
] [e E
4
3
2
1
=
(

0
0
(25)

From the above equation it follows that

E [
i
^
i
x ] x = (26)
which implies that the WLS estimate of each state variable has an expected
value equal to the true value.
}

= = dx ) (x f x ) (x E
Similarly we can find how close estimated measurement is to the true
measurement.
Consider (17) e ) W H G H I ( z z e
T 1
. .
= =

Then E
(
(
(
(
(
(

^
4
^
3
^
2
^
1
e
e
e
e
= E
(
(
(
(
(
(

^
4 4
^
3 3
^
2 2
^
1 1
z z
z z
z z
z z
= [ I - H G
-1
H
T
W ] E
(
(
(
(

4
3
2
1
e
e
e
e
=
(
(
(
(

0
0
0
0
(27)

From the above

E [
^
j
z ] = E [ z
j
] = E [ z
j true
+ e
j
] = z
j true
(28)

Equations (26) and (28) state that the WLS estimates of the state variables
and the measured quantities, on the average, are equal to their true values -
an obviously desirable property, and the estimates are said to be unbiased.
The estimated values can be accepted as unbiased estimates of the true
values provided no bad measurement data are present.
TEST FOR BAD DATA
When the system model is correct and the measurements are accurate,
there is good reason to accept the state estimates calculated by the WLS
estimator. But if a measurement is grossly erroneous or bad, it should be
detected and then identified so that it can be removed from estimator
calculations. The statistical properties of the measurement errors facilitate
such detection and identification.

As seen from Eq. (27), each estimated measurement error,
^
j
e = ( z
j
-
^
j
z ) is a
Gaussian random variable with zero mean. A formula for the variance of
^
j
e
can be determined from Eq. (17) namely
(17) e ) W H G H I (
e ) W H G ( H e ) X X ( H e X H e X H z z e
T 1
T 1

. . . .
=
= = + = =


in two steps.
Step 1

Using Eq. (17)

^
e
T
^
e = (z -
^
z ) (z -
^
z )
T
= [ I - H G
-1
H
T
R
-1
] e e
T
[ I - R
-1
H

G
-1
H
T
] (29)

It is to be noted that

1. LHS of Eq. (29) is a square matrix.
2. The two matrices on the RHS contain only constant elements and are
transpose of one another.
3. As seen in Eq. (22) E [ e e
T
] = R
Step 2

We take the expected value of both sides of Eq. (29)
E [
^
e
T
^
e ] = [ I - H G
-1
H
T
R
-1
] E [ e e
T
] [ I - R
-1
H

G
-1
H
T
]

= [ I - H G
-1
H
T
R
-1
] R [ I - R
-1
H

G
-1
H
T
]

= [ I - H G
-1
H
T
R
-1
] [ R - H G
-1
H
T
]

= [ I - H G
-1
H
T
R
-1
] [ R - H G
-1
H
T
R
-1
R ]

= [ I - H G
-1
H
T
R
-1
] [ I - H G
-1
H
T
R
-1
] R (30)

It can be verified that the matrix [ I - H G
-1
H
T
R
-1
] multiplied by itself
remains unaltered and so
E [
^
e
T
^
e ] = [ I - H G
-1
H
T
R
-1
] R

= R - H G
-1
H
T
(31)

The square matrix
^
e
T
^
e takes the form of Eq. (21) with typical entries given
by
^
j
^
i
e e . Therefore, substituting for
^
j
e = ( z
j
-
^
j
z ) in the diagonal entries,
we obtain
E [
2
^
j
e ] = E [ (z
j
-
^
j
z )
2
] = R

j j
(32)
where
R

j j
is symbol for the jth diagonal element of the matrix R

= R - H G
-1
H
T
.
Eq. (32) is actually the formula for the variance of
^
j
e , which makes j j
'
R
the standard deviation. Dividing each side of Eq. (32) by the number R

j j

gives

E
(
(
(

j j
'
2
^
j
R
e
= 1 (33)
E
(
(
(

j j
'
2
^
j
R
e
= 1 (33)

(34)
R
e
that interpreat can we above, the From
j j
'
^
j
is a STANDARD GAUSSIAN RANDAM VARIABLE with zero mean and
variance equals to 1.
Matrices R, given by R =
(
(
(
(
(

2
4
2
3
2
2
2
1
0 0 0
0 0 0
0 0 0
0 0 0
is the covariance matrix while
R

= R - H G
-1
H
T
is known as the modified covariance matrix.
Also from Eq. (15) namely e W H G x x
T 1
^

= we get
( x -
^
x ) ( x -
^
x )
T
= G
-1
H
T
R
-1
e e
T
R
-1
H G
-1
. Therefore,
E [ ( x -
^
x ) ( x -
^
x )
T
] = G
-1
H
T
R
-1
R R
-1
H G
-1

= G
-1
H
T
R
-1
H G
-1


= G
-1
= (H
T
R
-1
H)
-1
(35)
Now let us see how to detect the presence of bad data.

It is to be recalled that true measurement error e
j
, given by z
j
- z
j true
, is
never known in engineering applications. The best that can be done is to
calculate the estimated error
^
j
e , which then replaces e
j
in the objective
function. Accordingly, objective function value is obtained as
(36)

e
e w f
m m
N
1 j
2
j
2
^
j
2
N
1 j
^
j j
^

= =
= =
where N
m
is the number of measurements and the weighting factor w
j
is set
equal to 1/
j
2
. This weighted sum of squares of estimated errors,
^
f itself is
a random variable which has a well known probability distribution with
values (areas) already available in tabulated form. In order to use those
tables, we need to know the mean value of
^
f which can be determined as
discussed below.
(36)

e
e w f
m m
N
1 j
2
j
2
^
j
2
N
1 j
^
j j
^

= =
= =
Multiplying the numerator and the denominator of each term in Eq. (36) by
the calculated variance R

j j
we get
(37)
R
e R
f
m
N
1 j
2
j j
'
2
j
^
j
j j
'
^

=
=
Taking the expected value on both sides we get
E [ ] f
^
= E
(
(
(

=
m
N
1 j
j j
'
2
^
j
2
j
j j
'
R
e

R
(38)
Use of Eq. (33), namely E
(
(
(

j j
'
2
^
j
R
e
= 1 in the above results in
E [ ] f
^
=

=
m
N
1 j
2
j
j j
'

R
(39)
(The above result will be more clear by expanding Eq. (38), taking N
m
= 4)
For the dc circuit considered in Example 1, it will be shown in Example 2,
that the right hand side of the above equation has the numerical value of 2
(= 4- 2) for the four measurements and two state variables. More generally,
the expected value of
^
f is always numerically equal to the number of
degrees of freedom; i.e.
E [ ] f
^
= N
m
- N
s
(40)
where N
m
is the number of measurements and N
s
is the number of state
variables.
Thus the mean value of
^
f is an integer which is equal to N
m
- N
s
. This
number is also called as redundancy of the measurement scheme.
Statistical theory shows that the weighted sum of squares of estimated
errors
^
f has Chi-square distribution
2
k,
where is a Greek letter called
Chi, k is the number of degrees of freedom and relates to the area under
the
2
k,
curve. The Chi-square distribution very closely matches the
standard Gaussian distribution when k is large (k > 30), which is often the
case in power system applications.
p( )
2
for k degree of freedom
2

k,

Area (1 - )
Area
The figure below shows the probability density function of
2
k,
for a
representative small value of k. As usual, the total area under the curve
equals to 1, but here it is not symmetrically distributed.










The area under the curve to the right of k,
2
in Fig. equals . which is the
probability that
^
f
exceeds
k,
2
. The remaining area under the curve is the
probability (1 - ) that the calculated value of the
^
f
with k degree of
freedom, will take on a value less than
k,
2
i.e.
Pr (
^
f <

k,
2
) = (1 - )
(41)
Based on this, the critical value of
^
f
can be determined using the tabulated
value of
k,
2
shown in the Table below.
Table 1: Critical values of
2
k,




k 0.05 0.025 0.01 0.005 k 0.05 0.025 0.01 0.005

1 3.84 5.02 6.64 7.88 11 19.68 21.92 24.73 26.76
2 5.99 7.38 9.21 10.60 12 21.03 23.34 26.22 28.30
3 7.82 9.35 11.35 12.84 13 22.36 24.74 27.69 29.82
4 9.49 11.14 13.28 14.86 14 23.69 26.12 29.14 31.32
5 11.07 12.83 15.09 16.75 15 25.00 27.49 30.58 32.80

6 12.59 14.45 16.81 18.55 16 26.30 28.85 32.00 34.27
7 14.07 16.01 18.48 20.28 17 27.59 30.19 33.41 35.72
8 15.51 17.54 20.09 21.96 18 28.87 31.53 34.81 37.16
9 16.92 19.02 21.67 23.59 19 30.14 32.85 36.19 38.58
10 18.31 20.48 23.21 25.19 20 31.41 34.17 37.57 40.00


For example, choosing = 0.01 and k = (N
m
- N
s
) = 2, we can conclude that
the calculated value of
^
f

is less than the critical value of 9.21 with a
probability of (1 - 0.01) or 99% confidence since
2
0.01 2,
in Table.
The statistical properties of the measurement errors facilitate detection and
identification of bad measurement data.

The Chi-square distribution of
.
f provides a test for detection of bad
measurement. The procedure is as follows:

- Use the raw measurements z from the system to determine the WLS
estimates x of the system state from

z W H G x
T 1
.
=

- Substitute the estimates
.
x in equation
. .
= x H z .

- Calculated the estimated errors
. .
=
j j j
z z e
- Evaluate the sum of weighted squared estimated errors from



=
.
=
. .
= =
m m
N
1 j
2
j
2
j
N
1 j
2
j j

e
e w f (29)


- For the appropriate number of degrees of freedom
s m
N N k = and
a specified probability , find the critical Chi-square value
2
k,

from Table 1. Determine whether or not

.
f <
2
k,
(30)

is satisfied. If it is, then the measured raw data and the estimated
values of the state variables are accepted as being accurate.
- When the requirement of inequality (30) is not met, there is reason to
suspect the presence of at least one bad measurement. In that case
compute the diagonal elements of modified covariance matrix R

given by

R ] R H G H I [ R R H G H R H G H R R
1 T 1 -1 T 1 T 1 '
= = = (31)
and designate them as
'
j j
R .

It is to be noted that matrix R

is independent of measurements.

- Omit the measurement corresponding to the largest standardized
error, (considering only the magnitudes) namely
'
j j
j
R
e
.
(32)

and reevaluate the state estimates along with the sum of weighted
squared errors
.
f .
- If the new value of
.
f satisfies the Chi-square test of inequality (30),
then the omitted measurement has been successfully identified as
the bad data.
Example 2

Suppose that the weighting factors
4 1
w to w in Example 1 are the
corresponding reciprocals of error variances for the four meters of Fig,1.
Show that the expected value of
.
f is equal to the degree of freedom.

Solution

Expected value of
.
f is given by

=
.
=
m
N
1 j
2
j
'
j j

R
] f [ E

First let us evaluate the diagonal elements of

T 1 '
H G H R R

=

To take advantage of previous calculations in Example 1, we write the
matrix
'
R in the form

R ] R H G H I [ R
1 T 1 '
=
Then from Example 1, taking value of
1 T 1
R H G

we can compute

1 T 1
R H G H

=
(

(
(
(
(

0.4386 0.2281 1.2982 0.0351
0.2281 0.4386 0.0351 1.2982
0.375 0.125
0.125 0.375
0.625 0.125
0.125 0.625


Only the diagonal elements resulting from this equations are required or
calculating ] f [ E
.
.

1 T 1
R H G H

=
(
(
(
(

0.1930 * * *
* 0.1930 * *
* * 0.8070 *
* * * 0.8070


(
(
(
(
(

(
(
(
(

(
(
(
(

=
(
(
(
(
(

2
4
2
3
2
2
2
1
'
44
'
33
'
22
'
11
. . .
. . .
. . .
. . .
0.1930 * * *
* 0.1930 * *
* * 0.8070 *
* * * 0.8070
1 . . .
. 1 . .
. . 1 .
. . . 1
R . . .
. R . .
. . R .
. . . R
31 eq. From ,


The expected value of
.
f is

=
.
=
4
1 j
2
j
'
j j

R
] f [ E

=
2
4
2
4
2
3
2
3
2
2
2
2
2
1
2
1

) 0.193 1 (

) 0.193 1 (

) 0.807 1 (

) 0.807 1 (
+

= ( 1 + 1 + 1 + 1 ) ( 0.807 + 0.807 + 0.193 + 0.193 ) = 4 2 = 2

which is the number of degrees of freedom when the system of Example 1
has 4 measurements and 2 state variables.
Example 3

Using the Chi-square test of inequality
.
f <
2
k,
check for the presence
of bad data in raw measurements of Example 1. Choose = 0.01.

Solution

2 N and 4 N
s m
= = . Then k = 4 2 = 2
With k = 2 and = 0.01, Chi-square critical value
2
0.01 2,
as given in Table
1 is 9.21.

In Example 1,
(
(
(
(

=
(
(
(
(
(
(

.
.
.
.
0.00070
0.02596
0.00456
0.00875
e
e
e
e
4
3
2
1


Estimated sum of squares
.
f , is calculated as

=
.
.
=
4
1 j
2
j
2
j

e
f =
2
4
2
3
2
2
2
1
e 50 e 50 e 100 e 100
. . . .
+ + +

=
2 2 2 2
) 0.00070 ( 50 ) 0.02596 ( 50 ) 0.00456 ( 100 ) 0.00877 ( 100 + + +

= 0.043507

This value is less than the Chi-square critical value of 9.21. Therefore, we

conclude that the raw measurement set of Example 1 has no bad

measurement.
Example 4

Suppose that the raw measurement set for the system of Fig.1 is given by

(
(
(
(

=
(
(
(
(

V 4.40
V 6.98
A 3.02
A 9.01
z
z
z
z
4
3
2
1


Check for the presence of bad data using Chi-square test for = 0.01.
Eliminate any bad data detected and calculate the resultant state estimate
from the reduced data set.
Solution
For the given set of measurement we already calculated the estimated
errors as
(
(
(
(

=
(
(
(
(
(
(

.
.
.
.
0.49298 -
0.05965
0.15439
0.06228
e
e
e
e
4
3
2
1

Therefore
.
f =
2
4
2
3
2
2
2
1
e 50 e 50 e 100 e 100
. . . .
+ + +

=
2 2 2 2
) 0.49298 ( 50 ) 0.05965 ( 50 ) 0.15439 ( 100 ) 0.06228 ( 100 + + +
= 0.3879 + 2.3836 + 0.1779 + 12.1515 = 15.1009
This value of
.
f exceeds
2
0.01 2,
value of 9.21 and so we conclude that there
is at least one bad measurement. The standardized error estimates are
next calculated using the diagonal elements
'
j j
R as follows:
Matrix R

= [ I H G
-1
H
T
R
-1
] R

Then taking value of
1 T 1
R H G

from Example 1, we can compute

1 T 1
R H G H

=
(

(
(
(
(

0.4386 0.2281 1.2982 0.0351
0.2281 0.4386 0.0351 1.2982
0.375 0.125
0.125 0.375
0.625 0.125
0.125 0.625


Only the diagonal elements resulting from this equations are required,

diag. elements of
1 T 1
R H G H

=
(
(
(
(

0.1930 * * *
* 0.1930 * *
* * 0.8070 *
* * * 0.8070

From eq. 31, knowing R

= [ I H G
-1
H
T
R
-1
] R


(
(
(
(
(

(
(
(
(

(
(
(
(

=
(
(
(
(
(

2
4
2
3
2
2
2
1
'
44
'
33
'
22
'
11
. . .
. . .
. . .
. . .
0.1930 * * *
* 0.1930 * *
* * 0.8070 *
* * * 0.8070
1 . . .
. 1 . .
. . 1 .
. . . 1
R . . .
. R . .
. . R .
. . . R



R

=
(
(
(
(
(
(
(
(

50
0.193 - 1
. . .
.
50
0.1930 - 1
. .
. .
100
0.807 - 1
.
. . .
100
0.807 - 1
;
'
R =
(
(
(
(

0.1270 . . .
. 0.1270 . .
. . 0.04393 .
. . . 0.04393

Standardized estimated errors are computed as
3.8804
0.127
0.49298
R
e
0.4697
0.127
0.05965
R
e
3.5145
0.04393
0.15439
R
e
1.4177
0.04393
0.06228
R
e
'
44
4
'
33
3
'
22
2
'
11
1
=

=
= =
= =
= =
.
.
.
.

The magnitude of the largest standardized error corresponds to
measurement
4
z in this case. Therefore we identify
4
z as the bad
measurement and omit it, from the state estimation calculations. With the
three remaining measurements
. .
e hence and x are now computed.
H =
(
(
(

0.125 0.375
0.625 0.125
0.125 0.625

H
50
100
100
0.125 0.625 0.125
0.375 0.125 0.625
H R H G
1 T
(
(
(

= =


=
(

=
(
(
(

41.40625 13.28125
13.28125 47.65625
0.125 0.375
0.625 0.125
0.125 0.625
6.25 62.5 12.5
18.75 12.5 62.5

z R H G x
1 T 1
.
=
=
(

0.026522 0.007391
0.007391 0.023043

(

6.25 62.5 12.5
18.75 12.5 62.5
z
=
(

=
(
(
(

V 8.0265
V 16.0074
6.98
3.02
9.01
0.30435 1.56522 0.13044
0.477826 0.17391 1.34783

This gives
(
(
(

=
(

(
(
(

= =
. .
A 7.0061
A 3.0157
A 9.0013
8.0265
16.0074
0.125 0.375
0.625 0.125
0.125 0.625
X H Z
Thus
(
(
(

=
(
(
(

(
(
(

= =
. .
0.0261
0.0043
0.0087
7.0061
3.0157
9.0013
6.98
3.02
9.01
Z Z e
Sum of squares
.
f is calculated as

=
.
.
=
3
1 j
2
j
2
j

e
f =
2
3
2
2
2
1
e 50 e 100 e 100
. . .
+ +
= 0.0435 ) 0.0261 ( 50 ) 0.0043 ( 100 ) 0.0087 ( 100
2 2 2
= + +
Corresponding to k = 1 and = 0.01, Chi-square critical value is 6.64. Thus
.
f <
2
k,
. This means that the calculated state of
V 8.0265 x V, 16.0074 x
2 1
= = is quite acceptable. The new state estimate
based on the three good measurements essentially match those of
Example 1. ( V 8.0261 x V, 16.0072 x
2 1
= = )
+

Fig. 1

V
2
V
1
1

1

1
1

1

x
1 x
2
x
3
+
+

+

-

-
-

-

z
1
z
2
z
3
z
4
PROBLEM SET 1
1. Consider the circuit shown in Fig. 1, in which three loop current variables are
identified as x
1
, x
2
and x
3
. Measurement weights are w
1
= 100, w
2
= 100, w
3
= 50
and w
4
= 50. Meter readings are z
1
= 9.01 A, z
2
= 3.02 A, z
3
= 6.98 V and z
4
= 5.01 V.
Determine the weighted least square estimate of the three loop currents.
Determine the source voltages V
1
and V2 using (i) estimated loop currents (ii)
estimated measurements.






A A

V
V
2 1
+

Fig. 1

V
2
V
1
1

1

1
1

1

x
1 x
2
x
3
+
+

+

-

-
-

-

z
1
z
2
z
3
z
4








2. In the circuit shown in Fig. 1, consider the voltages at nodes 1 and 2 as state
variables. Measurements and measurement weights are same as in problem 1.
Determine the weighted least square estimates of the nodal voltages. Also
calculate the estimated measurements and estimated errors.
A A

V
V
2 1
Fig. 2

1

3

3

1

3

-

+

+

+

+

+

-

-

-

-

I
2 I
1
3. Five ammeters numbered A
1
to A
5
are used in the dc circuit shown in Fig. 2 to
determine the two unknown sources I
1
and I
2
. The standard deviations of meter
errors are 0.2 A for meters A
2
and A
5
. and 0.1 A for the other three meters. The
readings of the five meters are: 0.12 A, 1.18 A, 3.7 A, 0.81 A and 7.1 A
respectively.






a) Determine the weighted least square estimate of the source currents I
1
and I
2
.
b) Using the Chi-square test for = 0.01, identify the bad measurement, if any.
c) Find the WLS estimates of the source currents using reduced data set and
check if the result is statistically acceptable.
A
1

1
2 3
A
2
A
5
A
3
A
4
4. Redo Problem 3 when the unknowns to be determined are not the source
currents but, rather, the voltages at the nodes 1, 2 and 3 in Fig. 2.
5. Consider the circuit of Fig. 2 for which accuracy of ammeters and their
readings are the same as those specified in Problem 3. It is required to estimate
voltages at nodes 1, 2 and 3 treating those node voltages as state variables for
the following cases.
a) Suppose that meters A
4
and A
5
are found to be out of order, and therefore
only three measurements z
1
= 0.12 A, z
2
= 1.18 A and z
3
= 3.7 A are available.
Determine WLS estimate of the nodal voltages and the estimated errors.
b) Suppose meters A
2
and A
5
are out of order and measurements z
1
= 0.12 A, z
3
= 3.7 A and z
4
= 0.81 A, can the nodal voltages be estimated? Explain why by
examining the gain matrix.
6. Suppose that the two voltage sources of circuit shown in Fig. 1 are replaced
with new ones, and the meter readings are z
1
= 2.9 A, z
2
= 10.2 A, z
3
= 5.1 V and z
4

= 7.2 V. Measurement weights are: w
1
= 100, w
2
= 100, w
3
= 50 and w
4
= 50.
a) Taking the source voltages as state variables, determine their WLS estimates.
b) Using the Chi-square test for = 0.005, detect bad data.
c) Eliminate the bad data and determine again the WLS estimates of the source
voltages.
d) Check the result in part c again using Chi-square test.
40 MW
60 MW

70 MW

50 MW

Fig. 3

7. Five wattmeters are installed on the four-bus system of Fig. 3 to measure line
real power flows, where per-unit reactances of the lines are X
12
= 0.05, X
13
= 0.1,
X
23
= 0.04, X
24
= 0.0625 and X
34
= 0.08. Suppose that the meter readings show that
z
1
= P
12
= 0.34 p.u., z
2
= P
13
= 0.26 p.u., z
3
= P
23
= 0.17 p.u., z
4
= P
24
= - 0.24 p.u. and
z
5
= P
34
= - 0.22 p.u. where the variances of measurement error of all the meters
are given by
i
2
= 0.01
2
.






z
1
z
5
z
4
z
3
z
2
1
G
G
2
3 4
a) Apply the dc power flow method with bus 1 as reference and determine the
measurement Jacobian matrix H. Then compute WLS estimate of the phase
angles of the bus voltages in radians.
b) Using the Chi-square test for = 0.01, identify two bad measurements. If both
bad measurements are eliminated simultaneously, would it be possible to
estimate the states of the system?
c) Eliminate one of the bad measurements identified in part b and determine the
WLS estimates of the phase angles of the bus voltages using the reduced data
set. Do the same for the other bad measurement.
ANSWERS
1.
(
(
(

^
3
^
2
^
1
x
x
x
=
(
(
(

A 2.0100
A 3.0133
A 9.0033
; V
1
= 15.9966 V V
2
= 8.0366 V
2.
(

2
1
V
V
=
(

V 5.0107
V 7.0060
; z
^
=
(
(
(
(

5.0107
7.0060
3.0154
9.0013
; e
^
=
(
(
(
(

0.0007 -
0.0260 -
0.0046
0.0087

3.
(

^
2
^
1
I
I
=
(

A 8.0451
A 3.7218
; z
4
is identified as bad measurement.
(

^
2
^
1
I
I
=
(

A 8.2276
A 3.7576

4.
(
(
(

3
2
1
V
V
V
=
(
(
(

V 7.0375
V 2.9351
V 3.6595
; No bad data
5.
(
(
(

3
2
1
V
V
V
=
(
(
(

V 6.88
V 3.34
V 3.70
; Node 3 is not represented and hence node voltages cannot be
estimated.
6.
(

2
1
V
V
=
(

V 17.6649
V 8.0018
; z
2
is identified as the bad measurement.

(

2
1
V
V
=
(

V 16.5909
V 7.9727
; No bad data.
7. H =
(
(
(
(
(
(

12.5 12.5 0
16 0 16
0 25 25
0 10 0
0 0 20
;
(
(
(

4
3
2

=
(
(
(

rad 0.0040 -
rad 0.0240 -
rad 0.0175 -

Note that the standardized errors for the 4
th
and 5
th
measurements are equally
bad. If both are discarded, bus 4 will be virtually be disconnected from the
system, making state estimation impossible, The elimination of both z
4
and z
5
is
equivalent to deleting the 4
th
and 5
th
rows from H. it is easy to check that the
resulting gain matrix cannot be inverted.
By deleting z
4
:
(
(
(

4
3
2

=
(
(
(

rad 0.0068 -
rad 0.0244 -
rad 0.0174 -

By deleting z
5
:
(
(
(

4
3
2

=
(
(
(

rad 0.0024 -
rad 0.0244 -
rad 0.0174 -