Power System Operation And Control

Chapter Four-one
State Estimation
Yoseph Mekonnen (PhD)

Page 1
 Introductions
 Selective monitoring of the generation and transmission
system has been providing the data needed for
economic dispatch and load frequency control.

 More recently, however, interconnected power networks

have become more complex and the task of securely
operating the system has become more difficult.

 To help avoid major system failures and regional power

blackouts, electric utilities have installed more
extensive supervisory control and data acquisition
(SCADA) throughout the network to support computer
based systems at the energy control center.

Page 2
 …Contd..
 The data bank created is intended for a number of
application programs-some to ensure economic system
operation and others to assess how secure the system
would be if equipment failures and transmission-line
outages were to occur.
 Before any security assessment can be made or control
actions taken, a reliable estimate of the existing state
of the system must be determined.
 For this purpose the number of physical measurements
cannot be restricted to only those quantities required
to support conventional power-flow calculations.
 The inputs to the conventional power-flow program are
confined to the P, Q injections at load buses and P, |V|
values at voltage-controlled buses.
Page 3
 …Contd..
If even one of these inputs is unavailable, the conventional
power-flow solution cannot be obtained.

Moreover, gross errors in one or more of the input

quantities can cause the power-flow results to become
useless.

In practice, other conveniently measured quantities such

as P, Q line flows are available, but they cannot be used in
conventional power-flow calculations.

These limitations can be removed by state estimation

based on weighted least-squares calculations.

Page 4
 …Contd..
The techniques we will see provide an estimate of the
system state and a quantitative measure of how good the
estimate is before it is used for real-time power-flow
calculations or on-line system security assessment.
Besides the inputs required for conventional power-flow
analysis, additional measurements are made which usually
include megawatt and megavar flow in the transmission
lines of the system.
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.
Thus, gross errors detected in the course of state
estimation are automatically filtered out.
Page 5
 THE METHOD OF LEAST SQUARES
The electric power transmission system uses wattcmeters,
varmeters, voltmeters, and current meters to measure real
power, reactive power, voltages, and currents, respectively.
These continuous or analog quantities are monitored by
current and potential transformers (or other equivalent
devices) installed on the lines and on transformers and
buses of the power plants and substations of the system.
The analog quantities pass through transducers and
analog-to-digital converters, and the digital outputs are
then telemetered to the energy control center over various
communication links.
The data received at the energy control center is
processed by computer to inform the system operators of
the present state of the system.
Page 6
 …Contd...
The acquired data always contains inaccuracies which are
unavoidable since physical measurements (as opposed to
numerical calculations) cannot be entirely free of random
errors or noise.
These errors can be quantified in a statistical sense and
the estimated values of the quantities being measured are
then either accepted as reasonable or rejected if certain
measures of accuracy are exceeded
Because of noise, the true values of physical quantities are
never known and we have to consider how to calculate the
best possible estimates of the . unknown quantities.
The method of least squares is often used to "best fit"
measured data relating two or more quantities.

Page 7
 …Contd..
The best estimates are chosen as those which minimize
the weighted sum of the squares of the measurement
errors.
Consider the simple dc circuit of Fig. with five resistances
of 1Ω each and two voltage sources V1 and V2 of unknown
values which are to be estimated. The measurement set
consists of ammeter readings Z1 and Z2 and voltmeter
readings Z3 and Z4.

Page 8
 …Contd..
The symbol z is normally used for measurements
regardless of the physical quantity being measured, and
likewise, the symbol x applies to quantities being estimated.

The system model based on elementary circuit analysis

expresses the true values of the measured quantities in
terms of the network parameters and the true (but
unknown) source voltages x1 = V1 and x2 = V2•

Then, measurement equations characterizing the meter

readings are found by adding error terms to the system
model.

Page 9
 …Contd..
Using Nodal Analysis

The numerical coefficients are determined by the circuit

resistances and the terms el, e2, e3, and e4 represent
errors(residuals) in measuring the two currents Z1 and Z2 and
the two voltages Z3 and Z4.
Page 10
 …Contd..
If e1, e2, e3 and e4 were zero (the ideal case), then any two
of the meter reading would give exact and consistent
readings from which the true values Xl and X2 of V1 and V2
could be determined.
But in any measurement scheme there are unknown errors
which generally follow a statistical pattern.
Labeling the coefficients of Eqs, we obtain,

Page 11
 …Contd..
where Zj,true denotes the true value of the measured
quantity rearrange into the vector-matrix form:

Page 12
 …Contd..
In more compact notation:

which represents the errors between the actual

measurements z and the true (but unknown) values Ztrue=Hx
of the measured quantities.
The true values of xl and x2 cannot be determined, but we
can calculate estimates xˆ1 and xˆ2 , as we shall soon see.
Substituting these estimates gives estimated values of
the errors in the form:

Page 13
 …Contd..

Quantities with hats, are estimates of the corresponding

quantities without hats. It can be written:

Based on:

Where z=e+Hx

Page 14
 …Contd..
We must now decide upon a criterion for calculating the
estimates x1 and x2 from which:

are to be computed.

It is not desirable to choose the algebraic sum of the

errors to be minimized since positive and negative errors
could then offset one another and the estimates would not
necessarily be acceptable.

It is preferable to minimize the direct sum of the squares

of the errors.
Page 15
 …Contd..
However, to ensure that measurements from meters of
known greater accuracy are treated more favorably than
less accurate measurements, each term in the sum of
squares is multiplied by an appropriate weighting factor w
to give the objective function

We select the best estimates of the state variables as

those values x1 and x2 which cause the objective function f
to take on its minimum value.
According to the usual necessary conditions for minimizing
f, the estimates x1 and x2 are those values of X1 and X2
which satisfy the differential equation.

Page 16
 …Contd..
Necessary conditions

The notation |ˆx indicates that the equations have to be

evaluated from the state estimates ˆx = [ˆx1, ˆx2]T since
the true values of the states are not known.
The unknown actual errors ej are then replaced by
estimated errors ˆej, which can be calculated once the
state estimates ˆXi are known.
Page 17
 …Contd..
Equations in vector-matrix form become

Page 18
 …Contd..
where W is the diagonal matrix of weighting factors which
have special significance.
The partial derivatives to be constants given by the
elements of H, and so we obtain

Page 19
 …Contd..
Using the compact notation

In

Yealds

Solving for ˆx = [ˆx1, ˆx2]T

Page 20
 …Contd..
where ˆx1, ˆx2 are the weighted least-squares estimates
of the state variables.
Because H is rectangular, the symmetrical matrix HTWH
(often called the gain matrix G) must be inverted as a
single entity to yield G-1= (HTWH)-1, which is also
symmetrical.
We expect the weighted least-squares procedure to yield
estimates ˆxi which are close to the true values xi of the
state variables.
An expression for the differences (ˆxi-xi ) is found by
substituting for z = Hx + e into:

Page 21
 …Contd..
Yields

Canceling G with G-1 and rearranging the result lead to the

equation.

It is useful to check the dimensions of each term in the

matrix product above which is important for developing
properties of the weighted least squares estimation.

Page 22
 …Contd..
For the example circuit the matrix G-1HTW has the
overall row X column dimensions 2 X 4, which means that
anyone or more of the four errors el, e2, e3 and e4 can
influence the difference between each state estimate and
its true value.
In other words, the weighted least-squares calculation
spreads the effect of the error in any one measurement to
some or all the estimates; this characteristic is the basis
of the test for detecting bad data.
In a quite similar manner, we can compare the calculated
values ˆz = Hˆx of the measured quantities with their
actual measurements z by substituting for ˆx -x from

Page 23
 …Contd..
Into

Results
where I is the unit or identity matrix.

Immediate use of the above three equations is not

possible without knowing the actual errors ei, which is
never the practical case.

But we can use those equations for analytical purposes

after we have introduced the statistical properties of the
errors.

Page 24
 Example
In the dc circuit of Fig. below the meter readings are Z1 =
9.01 A, Z2 = 3.02 A, z3 = 6.98 V, and Z4 = 5.01 V. Assuming
that the ammeters are more accurate than the voltmeters,
let us assign the measurement weights wl = 100, w2 = 100,
w3 = 50, and W4 = 50, respectively.

Page 25
 …Contd..
Using Circuit analysis

Forming the H matrix

Page 26
 …Contd..
Results

Page 27
 …Contd..
Using this result to evaluate the symmetrical gain matrix
G gives.

Substituting numerical value

Page 28
 …Contd..
Becoming

which are the estimates of the voltage sources VI and V2.

Page 29
 …Contd..
The estimated measurements are calculated from ˆZ =
ˆHx as follows

the estimated errors in the measurements are then given by

Page 30
 …Contd..
The criterion for accepting or rejecting an estimate is
reasonable and if a grossly erroneous meter reading is
present and also to identify the bad measurement is found
by attaching statistical meaning to the measurement errors
in the least-squares calculations to develop aquantifiable
level of confidence

Page 31
Page 32