See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/324782095

Power System Dynamic State Estimation using Kalman Filtering Techniques

Thesis · July 2015

DOI: 10.13140/RG.2.2.31413.37608

CITATIONS READS
0 1,053

1 author:

Rampelli Manojkumar
BVRIT Hyderabad College of Engineering for Women
23 PUBLICATIONS 104 CITATIONS

SEE PROFILE

All content following this page was uploaded by Rampelli Manojkumar on 26 April 2018.

The user has requested enhancement of the downloaded file.

 POWER SYSTEM DYNAMIC STATE
ESTIMATION USING KALMAN FILTERING
TECHNIQUES

Thesis

Submitted in partial fulfilment of the requirements for the degree of

MASTER OF TECHNOLOGY in
POWER & ENERGY SYSTEMS
By
RAMPELLI MANOJ KUMAR
(13PS19F)

DEPARTMENT OF ELECTRICAL & ELECTRONICS ENGINEERING

NATIONAL INSTITUTE OF TECHNOLOGY KARNATAKA
SURATHKAL, MANGALORE -575025
JULY 2015
 POWER SYSTEM DYNAMIC STATE
ESTIMATION USING KALMAN FILTERING
TECHNIQUES
Thesis

Submitted in partial fulfilment of the requirements for the degree of

MASTER OF TECHNOLOGY in
POWER & ENERGY SYSTEMS
By
RAMPELLI MANOJ KUMAR
(13PS19F)
Under the guidance of
Dr. DEBASHISHA JENA
Assistant professor

DEPARTMENT OF ELECTRICAL & ELECTRONICS ENGINEERING

NATIONAL INSTITUTE OF TECHNOLOGY KARNATAKA
SURATHKAL, MANGALORE -575025
JULY 2015
 DECLARATION
By the P.G (M.Tech) student

I hereby declare that the Report of the P.G. Project Work entitled “POWER SYSTEM
DYNAMIC STATE ESTIMATION USING KALMAN FILTERING TECHNIQUES”
which is being submitted to the National Institute of Technology Karnataka Surathkal, in
partial fulfilment of the requirements for the award of the Degree of Master of Technology in
the department of Electrical and Electronics Engineering, is a bonafide report of the work carried
out by me. The material contained in this report has not been submitted to any University or
Institution for the award of any degree.

…………………………………………………………
(Register Number, Name and Signature of the student)
Department of Electrical and Electronics Engineering
National Institute of Technology Karnataka, Surathkal

Place: NITK, SURATHKAL

Date:
 CERTIFICATE

This is to certify that the Project Work Report entitled “Power System Dynamic State
Estimation Using Kalman Filtering Techniques” submitted by Mr. RAMPELLI MANOJ
KUMAR, Registration Number: 13PS19F in partial fulfillment of the requirements for the award
of degree of the MASTER OF TECHNOLOGY in POWER AND ENERGY SYSTEMS in the
Department of Electrical and Electronics Engineering is accepted as the record of the work
carried out by him.

Dr. DEBASHISHA JENA

Guide

CHAIRMAN DPGC/HOD.
Prof. JORA M. GONDA
 Acknowledgment

First, I thank GOD Almighty for giving me all the blessings in this venture and for helping
me in finishing project successfully.

I would like to express my sincere gratitude towards my guide Dr. DEBASHISHA JENA,
Department of Electrical and Electronics Engineering, NITK, without whose guidance and
support this work would have been impossible. I thank him for giving me an opportunity to
work under his guidance which is invaluable. His keen insight into the subject has helped me
in solving many problems during this work. His unflinching support, suggestions, directions
and feedbacks have helped in smooth progress of the project work. He has been a constant
source of inspiration in all possible ways for successful completion of my project.

I am extremely grateful to my HOD Mr. Jora M Gonda, Department of Electrical and

Electronics Engineering, NITK, Surathkal, for his encouragement and providing me with
sufficient computational facilities to successfully complete the project work.

I express my heartfelt thanks to all the teaching and non-teaching staff of the Department
of Electrical and Electronics Engineering for full co-operation and assistance. I also extend
thanks to the librarian and library staff members.

It’s my pleasure to thank my Parents and my Family Members for the support,
encouragement and love they gave me. They are the one who kept me on high spirits in hard
times.

I thank all my Friends for their help, discussions, useful suggestions and who made my
stay in NITK, Surathkal.

Finally, I thank all those who helped me directly or indirectly for the successful
completion of this thesis.

RAMPELLI MANOJ KUMAR

 ABSTRACT

The Kalman filter is a set of mathematical equations which is used to estimate the state of a
system, in a way that minimizes the mean of the squared error. In this project Kalman
filtering is used for the estimation of both static states i.e. Voltage and its angle at all buses
and dynamic states i.e. rotor angle w.r.t. synchronously rotating frame in rad and relative
angular speed in rad/sec of all the generators present in the system for an IEEE 14 bus power
system network. We mainly focus on the comparison of various kalman filtering
techniques.EKF and DKF algorithms have been compared by finding static state estimates. In
EKF we can consider nonlinear Process and measurement functions where it is limited to
only linear functions in case of DKF. Because of this limitation using DKF we can only
estimate one state of the power system but not both. Since the actual nonlinear equation has
to be approximated for applying DKF so that it becomes linear, it leads to the presence of
single state i.e. the angle of voltage at all buses of the network in the approximated equation.
Apart from this limitation estimated accuracy is also improved from DKF to EKF. Similarly
EKF and UKF algorithms are compared by finding dynamic state estimates. These two
algorithms are tested on an IEEE-14 bus,5-generator test system by applying test cases like
sudden load change and configuration topology error to show how adaptive these filters
during those conditions. The coding required for the implementation of algorithms has been
done through MATLAB.
TABLE OF CONTENTS

LIST OF FIGURES iii

LIST OF TABLES iv

LIST OF SYMBOLS v

LIST OF ACRONYMS vii

1 INTRODUCTION 1

1.1 BACKGROUND AND MOTIVATION 1

1.1.1 Applications of Power System State Estimation 2
1.2 TYPES OF STATE ESTIMATION 3
1.2.1 Static State Estimation 3
1.2.2 Tracking State Estimation 5
1.2.3 Dynamic State Estimation 6
1.3 OBJECTIVES 7
1.4 ORGANIZATION OF THESIS 7
2 LITERATURE SURVEY 8
3 KALMAN FILTER TECHNIQUES 11
3.1 INTRODUCTION 11
3.2 DISCRETE KALMAN FILTER-DKF 11
3.2.1 The Computational Origins of the Filter 13
3.2.2 Algorithm of DKF 14
3.3 EXTENDED KALMAN FILTER-EKF 15
3.3.1 Computational Origins of the Filter 16
3.4 UNSCENTED KALMAN FILTER-UKF 19
4 APPLICATION OF KALMAN FILTER TECHNIQUES IN POWER

SYSTEMS STATE ESTIMATION 21

4.1 POWER SYSTEM STATIC STATES ESTIMATION USING DKF AND EKF 21
4.1.1 Working Methodology 21
4.1.2 Static States Estimation Using DKF 22
4.1.3 Algorithm of DKF 23

i
 4.1.4 Static States Estimation Using EKF 23
4.1.5 Algorithm of EKF 24
4.2 POWER SYSTEM DYNAMIC STATES ESTIMATION USING EKF AND UKF 24
4.2.1 Working Methodology 24
4.2.2 Algorithm of Working Methodology 28
4.2.3 Dynamic States Estimation Using EKF 32
4.2.4 Algorithm of EKF 33
4.2.5 Dynamic States Estimation Using UKF 33
4.2.6 Algorithm of UKF 34
5 RESULTS AND DISCUSSION 35
5.1 POWER SYSTEM STATIC STATES ESTIMATION USING DKF AND EKF 35
5.1.1 Observations 37
5.2 POWER SYSTEM STATIC STATES ESTIMATION USING EKF AND UKF 37
5.2.1 Observations 43
6 CONCLUSIONS AND FUTURE SCOPE 45
6.1 CONCLUSIONS 45
6.2 FUTURE SCOPE 45

PUBLICATIONS 46

REFERENCES 47

APPENDIX I 50

APPENDIX II 54

ii
LIST OF FIGURES

3.1 Operation of the Kalman Filter 15

4.1 IEEE 5-generaor, 14-bus network 21
5.1 Calculated Power flows using DKF Algorithm 36
5.2 Calculated Power flows using EKF Algorithm 36
5.3 Variation of Rotor Angle of Generator 2 using EKF and UKF Algorithms during
Configuration Topology Error 39
5.4 Variation of Angular Speed of Generator 2 using EKF and UKF Algorithms during
Configuration Topology Error 40
5.5 Variation of Rotor Angle of Generator 3 using EKF and UKF Algorithms during
Configuration Topology Error 40
5.6 Variation of Angular Speed of Generator 3 using EKF and UKF Algorithms during
Configuration Topology Error 41
5.7 Variation of Rotor Angle of Generator 2 using EKF and UKF Algorithms during Sudden
Load Change 41
5.8 Variation of Relative Angular Speed of Generator 2 using EKF and UKF Algorithm
during Sudden Load Change 42
5.9 Variation of Rotor Angle of Generator 3 using EKF and UKF Algorithm during Sudden
Load Change 42
5.10 Variation of Relative Angular Speed of Generator 3 using EKF and UKF Algorithm
during Sudden Load Change 43
7.1 DC Ciruit Example for Weighted Least Squares Estimation 50

iii
LIST OF TABLES

5.1 Estimated and Measured Square Error DKF and EKF 36

5.2 Average Estimated Error of EKF and UKF 39
8.1 IEEE 5 generator,14 bus network line data 54
8.2 IEEE 5 generator,14 bus network generator dynamic data 55
8.3 IEEE 5 generator,14 bus network bus data 55

iv
LIST OF SYMBOLS

xk : Vector of state variables.

wk : Process noise

vk : Measurement noise

zk : Vector of Measurements

Q : Process noise covariance

R : Measurement noise covariance
K : Kalman Gain
xkpost : Posteriori State Estimate

xkpri : Priori State Estimate

Pkpost : Posterior estimate error covariance

Pkpri : Posterior estimate error covariance

H : Measurement function in terms of state variables

e xpri
k
: Prediction Error

e zpri
k
: Measurement residual

ekpri : Priori estimate error

ekpost : Posterior estimate error

H : Inertia Constant

M : Angular momentum

xd : Damping Reactance

J : Combined moment of inertia of prime mover and generator

m : Angular displacement of rotor w.r.t. stationary reference frame

m : Synchronous speed of Generator

e : Electrical power angle/Rotor angle w.r.t. synchronously rotating reference

Frame

v
m : Mechanical power angle

Vn : (ng  nb )  1 column vector which includes generator internal voltages and

voltages at all buses.

ng : Number of generators

nb : Number of buses

SL : Apparent Load Power

PL : Active Load Power

QL : Reactive Load Power

Eg : Generator Internal Voltages

Vb : Bus Voltages

ni : Total number of Instants

ns : Total number of States

 ij : Voltage angle between the buses i,j

bij : Susceptance between the buses i,j

N Li : Number of buses connected to i th bus

vi
LIST OF ACRONYMS

SSE : Static State Estimation

TSE : Tracking State Estimation
DSE : Dynamic State Estimation
DKF : Discrete kalman Filter
EKF : Extended Kalman Filter
UKF : Unscented Kalman Filter
DC : Direct Current

vii
CHAPTER 1

INTRODUCTION

State estimation is the process of assigning a value to an unknown system state

variable based on measurements from that system according to some criteria. Usually,
the process involves imperfect measurements that are redundant and the process of
estimating the system states is based on a statistical criterion that estimates the true
value of the state variables to minimize or maximize the selected criterion. A
commonly used and familiar criterion is that of minimizing the sum of the squares of
the differences between the estimated and true values of a function.

The ideas of least-squares estimation have been known and used since the
early part of the nineteenth century. The major developments in this area have taken
place in the twentieth century in applications in the aerospace field [27]. In these
developments, the basic problems have involved the location of an aerospace vehicle
(i.e. missile, airplane, or space vehicle) and the estimation of its trajectory given
redundant and imperfect measurements of its position and velocity vector. In many
applications, these measurements are based on optical observations and/or radar
signals that may be contaminated with noise and may contain system measurement
errors. State estimators may be both static and dynamic. Both types of estimators have
been developed for power systems.

1.1 BACKGROUND AND MOTIVATION

In a power system, the static state variables are the voltage magnitudes and relative
phase angles at the system nodes. We need measurements in order to estimate the
system performance in real time for both system security control and constraints on
economic dispatch. The inputs to an estimator are noisy power system measurements
of voltage magnitudes and power, VAR, or ampere-flow quantities. The estimator is
designed to produce the “best estimate” of the system voltage and phase angles,
recognizing that there are errors in the measured quantities and that there may be

1
redundant measurements. The output data is then used for the security-constrained
dispatch and control of the system.

By simply checking each measured value against its limit, the power system
operators can tell where problems exist in the transmission system so that they can
take actions to relieve overloaded lines or out-of-limit voltages. There are many
problems in monitoring a transmission system because of the nature of the
measurement transducers and from communications problems in transmitting the
measured values back to the operations control centre.

Transducers from power system measurements will be subject to errors. The

errors may go undetected if they are small and can cause misinterpretation by those
reading the measured values. Transducers may have gross measurement errors that
make their output useless in a case like if a transducer connected up backward it gives
the negative of the value being measured. Finally, the telemetry equipment often
experiences periods when communications channels are completely out which makes
the system operator not to know information about some part of the power system
network. It is for these reasons that power system state estimation techniques have
been developed. A state estimator, as we will see shortly, can “smooth out” small
random errors in meter readings, detect and identify gross measurement errors, and
“fill in” meter readings that have failed due to communications failures.

1.1.1 APPLICATIONS OF POWER SYSTEMS STATE

ESTIMATION
Information flow between the various functions is performed in an operations control
centre computer system. Remote terminal units which encode measurement
transducer outputs and opened/closed status information into digital signals that are
transmitted to the operations centre over communications circuits gives information
about the power system to the system. The control centre can transmit control
information such as raise/lower commands to generators and open/close commands to
circuit breakers and switches. The information coming into the control centre will be
broken down as breaker/switch status indications and analogue measurements. The
analogue measurements of generator output must be used directly by the AGC
program whereas all other data will be processed by the state estimator before being

2
used by other programs. We need to know how the transmission lines are connected
to the load and generation buses to run the state estimator which is known as the
network topology. A program must be provided which reads the telemeter
breaker/switch status indications and restructures the electrical model of the system
which is known as network topology program. The network topology program
describes about each substation and how the transmission lines are connected to the
substation equipment. Bus sections that are connected to other bus sections through
closed breakers or switches are designated as belonging to the same electrical bus.
The electrical model of the power system’s transmission system is sent to the
state estimator program together with the analogue measurements. The output of the
static state estimator consists of all bus voltage magnitudes and phase angles,
transmission line MW and MVAR flows calculated from the bus voltage magnitude
and phase angles, and bus loads and generations calculated from the line flows. These
quantities, together with the electrical model developed by the network topology
program, provide the basis for the economic dispatch program, contingency analysis
program, and generation corrective action program.
1.2 TYPES OF STATE ESTIMATION
There are mainly three types of State Estimation procedures as

1.2.1 Static State Estimation-SSE

1.2.2 Tracking State Estimation-TSE

1.2.3 Dynamic State Estimation-DSE

1.2.1 Static State Estimation-SSE

As the power system grows larger and becomes complex, real time monitoring and
controlling becomes much significant for reliable operation of the power system. The
energy management system functions are responsible for monitoring and control. So
an efficient and accurate state estimation is compulsory for an efficient and reliable
operation of the power system. The state estimation for a power network involves,
collecting the real time Measurement data, which includes line flows, injection
measurements and voltage measurements, through the SCADA and calculating the
state vector, using predefined State estimation algorithms. If the state vector is

3
obtained for an instant of time k, from the measurement set of the same instant of
time, then such an estimator is called the static state estimator. To know the state of
the power system regularly, this process of calculating the state vector is repeated at
suitable intervals of time. Static state estimators are widely used in power systems and
play a very important role for the reliable operation of the transmission and
distribution systems.

Need for Tracking State Estimation

Under normal conditions, power system is said to be a quasi-static system. But as the
power system expands, the system becomes extremely large for the state estimation to
be carried out at short intervals of time as it consumes heavy computing resources.
Hence a new technique called “Tracking State Estimation” was developed, wherein
the state estimate once calculated, is simply updated for the next instant of time, with
new set of measurement data obtained for that instant, instead of again running the
static state estimation algorithm fully. Tracking estimators help the EMS to keep track
of the continuously changing power system without actually performing the whole
state estimation. This allows continuous monitoring of the system, without excessive
usage of the computing resources.
Need for Dynamic State Estimation
The simplest way of following the changes in the system is tracking. But tracking
does not include any explicit physical modelling of the time behaviour of the system.
This leads to the ‘Dynamic State Estimators’ (DSE), where the actual physical
modelling of the time varying nature of the system is used. These algorithms have
dual advantages of being more accurate and possessing the ability to predict the state
of the system one step ahead. That is, from the knowledge of the state vector at an
instant of Time ‘t’, and the physical model of the system, the DSE predicts the state of
the power system at the next instant of time 't+1’. This forecasting ability has
tremendous advantages in performing security analysis and allows more time for the
operator to take control actions. Hence DSE algorithms form an important branch of
state estimation Techniques, with a potential to impact the very nature of operation of
the real time monitoring and control. Let us discuss TSE and DSE in detail.

4
1.2.2 Tracking State Estimation-TSE
As mentioned earlier, the power system is considered to be a quasi static system.
Hence the changes in the system occur slowly. It means that the state may not vary
much in a short span of time. But sometimes it becomes very important to have a
close monitoring of the system, like during the picking up of load by a generator or
during some contingency etc.
In order to have a real time monitoring system, state estimation must be
performed as and when every new data set arrives [7]. But as state estimation is
computationally heavy it becomes necessary to estimate states at such a high
frequency. But if the duration between two estimates is too large, it results in a week
co-relation between the estimated states, rendering the bad data detection very
difficult.
It can clearly be seen, that tracking estimation allows an easy update to the
existing state vector without much computationally complex operations. Tracking
estimation plays its role in between any two predefined execution instants of static
estimation. That is, the static state estimation will be performed in the control centre
at regular intervals or when there is sufficient change in the power system. But these
two instants are separated by a large amount of time. Tracking estimation helps the
control station to monitor the power system in between these two instants of time so
that we will get a real time update of the system without actually performing the
entire state estimation. The concept of tracking views the estimator as a digital
feedback loop which uses the new measurements to obtain new estimates, by
correcting an old estimate by a feedback gain signal, operating through a gain matrix.
Advantages of Tracking State Estimators
 Continuous tracking of the system helps the system operator to take better
decisions, in case of an emergency.
 Techniques proposed are computationally lighter and hence can be easily
implemented online
 Some of the tracking estimation techniques proposed, perform well even under
loss of information, bad data or ill conditioning hence make the system more
robust.

5
  If scanning of measurements takes longer time (especially in large networks)
sequential processing techniques can be used to obtain real time update of the
system.
1.2.3 Dynamic State Estimation-DSE
An easy way of following the changes in a power system on a real time basis is by
using the tracking state estimation techniques. But these techniques, though
computationally very efficient, do not use any physical modelling of the time varying
nature of the power system. Hence may not be as accurate as desirable. Lack of any
physical modelling also results in one of the main drawbacks of a tracking state
estimator, which is the lack of the ability to predict the state vector in advance.
Dynamic state estimation uses the present (and sometimes previous) state of the
power system along with the knowledge of the system’s physical model, to predict the
state vector for the next time instant [7]. This prediction feature of the DSE provides
vital advantages in system operation, control, and decision-making. It allows the
operator more time to act in cases of emergency, helps in detection of Anomalies, bad
data etc. Since DSE can play a major role in monitoring and risk prediction with their
unique ability of predicting the state vector one time stamp ahead, their association
with phasor measurement units is a great advantage for real time monitoring,
detection and control of power systems. From this point of view the PMU based DSE
techniques are of extreme importance for the modern day energy management
systems.
Advantages of Dynamic State Estimation
Because of the ability of predicting the state vector one step ahead
 It allows security analysis to be carried out in advance and hence allows the
operator to have more time during emergencies.
 It helps to identify and reject bad data and hence improves the estimator
performance.
 In cases where pseudo measurements are to be used, DSE readily provides
high quality values and hence avoids ill conditioning.

6
  DSE can be used for data validation, as the states are predicted one time stamp
before. Similarly, with the help of the predicted state vector we can identify
sudden changes in the system, topological errors and other anomalies.

1.3 OBJECTIVES

 To estimate the static states of IEEE 5-generator 14-bus network i.e. voltage
and its angle at all buses using Discrete Kalman Filter and Extended Kalman
Filtering Techniques
 To estimate the dynamic states of IEEE 5-generator 14-bus network i.e.
relative angular speed and rotor angle of generators using Extended Kalman
Filter and Unscented Kalman filtering Techniques
1.4 ORGANIZATION OF THESIS
 Chapter 2 presents the literature survey on various types the state estimation
techniques
 Chapter 3 presents the various types of Kalman Filtering techniques used for
state estimation purpose
 Chapter 4 presents the Application of Kalman Filtering techniques in power
systems state estimation.
 Chapter 5 presents the Results and Discussion regarding the above study.
 Chapter 6 presents Conclusions and Future Scope of work.

7
CHAPTER 2

LITERATURE SURVEY

It is very important to control and monitor the power system, because of the
integration of several renewable energy sources with it. So with this regard we
consider state estimation as the backbone of the control and monitoring of the system.
It was initially formulated by F. C. Schweppe. In [1], discussion covers the overall
problem, mathematical modelling, and general algorithms for state estimation,
detection, and identification. In [2], same procedure has been done by considering an
approximate mathematical model. In [3], the authors are discussed about various
implementation problems. In general there are three types of state estimation
techniques i.e. Static State Estimation (SSE), Tracking State Estimation (TSE) and
Dynamic State Estimation (DSE). In Static State Estimation we consider only one
measurement set at a point of time and the corresponding state is estimated which has
been discussed in above mentioned papers. In case of TSE the estimation starts from
the last estimated state variables rather than flat start as in SSE which has been
discussed in [4]. These two methods use only one step for estimation. To further
include the several advantages apart from estimating states a two step method of
estimation has been introduced which is DSE which has been discussed in [5-7]. Here
the extra step involved in the process is very much useful which provides several
advantages like allowing power system operator to get extra time for making control
decisions during emergency conditions. By considering that advantage Kalman filter
[8], exponential smoothing [9]–[11], [12]–[14] and ANN [15]–[18] have been
successfully used in the power system Forecasting Aided State Estimation context.

8
Forecasting-aided state estimation (FASE) covering a period of three decades of
intense research is presented [19]. The main achievements of FASE algorithms are
collected and discussed. Many topics of SE are focused: data redundancy, state
forecasting, observability analysis, innovation analysis, state filtering, bad data,
network configuration, and parameter error processing. Aspects of modeling,
forecasting techniques, and computational effort are also addressed. In contrast with
the large number of publications reporting research on FASE, to date, only [20]
describes the implementation experience of this approach in a European
(Czech&Slovak Energy Dispatch Center, Prague) control center. Thus, it is much
needed further research and development (R&D) into the practical use of estimators
with forecasting capability. ANN-based state forecasting model can be found in [21].
Skill in forecasting is generally defined as the accuracy of the forecasts obtained with
a particular method in relation to the accuracy of the forecasts based solely on a naive
forecasting method (the one in which the most recent observed value of the variable
of interest is used as one-step ahead forecasting). A naive method represents a
standard of reference, in the sense that it establishes the zero of the scale on which
skill at forecasting is evaluated. In order to assess the capability of the adopted
forecasting model for reproducing the behavior of system state, the mean absolute
scaled error (MASE) is considered [22].

We know that for accurate state estimation we need data redundancy which is
significant to find if there is bad data in measurement values. The main challenge of
DSE is that if there are any sudden changes in the load and generation, there is a large
transition in the state variables in the area which is close to the affected area. In that
case if you consider now at tth instant,

1) We need to predict the states at t+1th instant

2) Using the predicted state variables of instant ‘t’ which were calculated previous
instant and measurements at instant ‘t’ states at instant ‘t’ are estimated.

Suppose in between tth and t+1th instants if there is a sudden change in the
system, As we already predicted state variables at t+1th instant, those predicted values
don’t have the effect of sudden change. If we use those predicted states and the
present instant measurement set, surely we get erroneous state variables at the end of

9
correction step. If it happens the entire DSE performance will degrade subsequently
because at every instant state estimate depends on previously predicted state. To avoid
this we use innovation analysis where measurement set and predicted states will be
compared and decided whether sudden change has occurred or not in the system. In
1960, R.E. Kalman published his famous paper describing a recursive solution to the
discrete-data linear filtering problem [23]. The Discrete Kalman Filter (DKF)
addresses the general problem of trying to estimate the state x  R of a discrete time
controlled process that is governed by the linear stochastic difference equation where
as EKF and UKF addresses nonlinear stochastic difference equation. In a power
system network voltage and its angle are considered as static states. Rotor speed and
its angle are considered as dynamic states. In [24], an IEEE-14 bus test system is used
for finding static states using Kalman filtering algorithm. The various algorithms are
compared in [25] by considering power system stability problem as a case study. A
new state estimator by considering zero injection constraints has been developed in
[26] by UKF for finding static states. In this project we considered DKF and EKF for
state estimation process to find the static states and EKF and UKF for state estimation
process to find the dynamic states of IEEE 5 generator 14 bus system.

10
CHAPTER 3

KALMAN FILTERING TECHNIQUES

3.1 INTRODUCTION

Kalman filtering, is an algorithm which uses a series of measurements observed over

time, containing noise (random variations) and other inaccuracies, and produces
estimates of unknown variables more precisely than those based on a single
measurement alone. More formally, the Kalman filter operates recursively on noisy
input data to produce a statistically optimal estimate of the system state. The filter is
named after Rudolf (Rudy) E. Kalman, one of the primary developers of its theory.

As the Kalman filter uses a system's dynamics model (e.g., physical laws of
motion), known control inputs to that system, and multiple sequential measurements
(such as from sensors) to form an estimate of the system's varying quantities (its state)
that is better than the estimate obtained by using any one measurement alone, it is a
common sensor fusion and data fusion algorithm.

All measurements and calculations based on models are estimates to some

degree. Noisy sensor data, approximations in the equations that describe how a system
changes, external factors that are not accounted for will introduce some uncertainty
about the inferred values for a system's state. The Kalman filter averages a prediction
of a system's state with a new measurement using a weighted average. The purpose of
the weights is that values with better (i.e., smaller) estimated uncertainty is "trusted"
more. The weights are calculated from the covariance, a measure of the estimated
uncertainty of the prediction of the system's state. The result of the weighted average
is a new state estimate that lies between the predicted and measured state, and has a
better estimated uncertainty than either alone. This process is repeated every time

11
step, with the new estimate and its covariance informing the prediction used in the
following iteration. This means that the Kalman filter works recursively and requires
only the last "best guess", rather than the entire history, of a system's state to calculate
a new state.

Because the certainty of the measurements is often difficult to measure

precisely, it is common to discuss the filter's behaviour in terms of gain. The Kalman
gain is a function of the relative certainty of the measurements and current state
estimate, and can be "tuned" to achieve particular performance. With a high gain, the
filter place more weight on the measurements, and thus follows them more closely.
With a low gain, the filter follows the model predictions more closely, smoothing out
noise but decreasing the responsiveness. At the extremes, a gain of one causes the
filter to ignore the state estimate entirely, while a gain of zero causes the
measurements to be ignored.

When performing the actual calculations for the filter (as discussed below), the
state estimate and covariance are coded into matrices to handle the multiple
dimensions involved in a single set of calculations. This allows for representation of
linear relationships between different state variables (such as position, velocity, and
acceleration) in any of the transition models or covariances. There are various types of
Kalman filters. In this project Discrete Kalman Filter, Extended Kalman Filter and
Unscented Kalman Filter has been discussed.

3.2 DISCRETE KALMAN FILTER-DKF

The Kalman Filter addresses the general problem of trying to estimate the state x  R
of a discrete time controlled process that is governed by the linear stochastic
difference equation

xk  Ax k 1  Bu k 1  wk 1 (3.1)

with a measurement z  R that is

z k  Hx k  v k (3.2)

12
xk is the vector of state variables i.e. angles of voltages. wk and vk are process and

measurement noises respectively. wk and vk are assumed to be independent of each

other, white and with normal probability distributions.

p( w)  N (0, Q)
(3.3)
p(v)  N (0, R)

In general process noise covariance Q, measurement noise covariance R, Constant

matrices A and H might change with each time step, however here we assume they are
constant.

3.2.1 The Computational Origins of the Filter

Define a priori and posteriori estimate errors as

ekpri  x k  x kpri
(3.4)
ekpost  x k  x kpost

The a prior estimate error covariance is then

T
Pkpri  E (e kpri e kpri )
(3.5)

and the a posterior estimate error covariance is

T
Pkpost  E (ekpost Pe kpost ) (3.6)

In deriving the equations for the Kalman filter, we begin with the goal of finding an
equation that computes a posteriori state estimate as a linear combination of an a
priori estimate and a weighted difference between an actual measurement and a
measurement prediction as shown below in Equation (3.7).

xkpost  xkpri  K ( z k  Hx kpri ) (3.7)

The difference in Equation (3.7) is called the measurement innovation, or the

residual. The residual reflects the discrepancy between the predicted measurement
and the actual measurement. A residual of zero means that the two are in complete
agreement. The matrix K in Equation (3.7) is chosen to be the gain or blending factor
that minimizes the a posteriori error covariance Equation (3.6). This minimization can

13
be accomplished by first substituting Equation (3.7) into the above definition for ekpost

, substituting that into Equation (3.6), performing the indicated expectations, taking
the derivative of the trace of the result with respect to K, setting that result equal to
zero, and then solving for K. One form of the resulting K that minimizes Equation
(3.6) is given by

K k  Pkpri H T ( HPkpri H T  R) 1 (3.8)

Looking at Equation (3.8) we see that as the measurement error covariance
approaches zero, the gain K weights the residual more heavily. Specifically,

lim K k  H 1 (3.9)
Rk 0

On the other hand, as the a priori estimate error covariance approaches zero, the gain
K weights the residual less heavily. Specifically,

lim K k  0 (3.10)
Pkpri 0

All of the Kalman filter equations can be algebraically manipulated into to several
forms. Equation (1.8) represents the Kalman gain in one popular form.

Importance of Kalman Gain

As the measurement error covariance approaches zero, the actual measurement is

“trusted” more and more, while the predicted measurement is trusted less and less.
On the other hand, as the a priori estimate error covariance approaches zero the actual
measurement is trusted less and less, while the predicted measurement is trusted more
and more.

3.2.2 Algorithm of DKF

The Kalman filter estimates a process by using a form of feedback control the filter
estimates the process state at some time and then obtains feedback in the form of
(noisy) measurements. As such, the equations for the Kalman filter fall into two
groups: time update equations and measurement update equations. The time update
equations are responsible for projecting forward (in time) the Current state and error
covariance estimates to obtain the a priori estimates for the next time step. The
measurement update equations are responsible for the feedback—i.e. for

14
incorporating a new measurement into the a priori estimate to obtain an improved a
posterior estimate. The time update equations can also be thought of as predictor
equations, while the measurement update equations can be thought of as corrector
equations. Indeed the final estimation algorithm resembles that of a predictor-
corrector algorithm for solving numerical problems. The time update projects the
current state estimate ahead in time. The measurement update adjusts the projected
estimate by an actual measurement at that time.
Again notice how the time update equations project the state and covariance
estimates forward from time step to step and Bare from Equation (3.1), while is from
Equation (3.3). The first task during the measurement update is to compute the
Kalman gain. The next step is to actually measure the process to obtain, and then to
generate an a posteriori state estimate by incorporating the measurement as in
Equation (3.7). After each time and measurement update pair, the process is repeated
with the previous posterior estimates used to project or predict the new a priori
estimates. This recursive nature is one of the very appealing features of the Kalman
filter—it makes practical implementations much more feasible than (for example) an
implementation of a Wiener filter which is designed to operate on all of the data
directly for each estimate. The Kalman filter instead recursively conditions the current
estimate on all of the past measurements. Figure 3.1 shown below offers a complete
picture of the operation of the Kalman filter.

Time Measurement
update update
(Predict) (Correct)

Figure 3.1 Operation of the Kalman Filter

3.3 EXTENDED KALMAN FILTER-EKF

The Kalman Filter addresses the general problem of trying to estimate the state x  R
of a discrete time controlled process that is governed by the non-linear stochastic
difference equation

15
 xk  f ( xk 1 , uk 1 , wk 1 ) (3.11)

with a measurement z  R that is

z k  h( xk , vk ) (3.12)

where xk is the vector of state variables. wk and vk are process and measurement

noises respectively. wk and vk are assumed to be independent of each other, white

and with normal probability distributions.

In practice of course one does not know the individual values of the noise and at
each time step. However, one can approximate the state and measurement vector
without them as
~
xk  f ( xk 1 , u k 1 ,0)
~z  h( x , v ) (3.13)
k k k

3.3.1 The Computational Origins of the Filter

To estimate a process with non-linear difference and measurement relationships, we

begin by writing new governing equations that linearize an estimate about Equation
(3.13)

xk  ~
x  A( xk 1  xˆk 1 )  Wwk 1
(3.14)
z  ~z  H ( x  ~
k k x )  Vv
k k k

where

xk and zk are the actual state and measurement vectors,

~x and ~z are the approximate state and measurement vectors from (3.13),
k k

xkpost
1 is an a posterior estimate of the state at step k-1,

wk and vk represent the process and measurement noise.

A[i , j ] , W[i , j ] , H [i , j ] , V[ i , j ] are Jacobians defined as

16
 f [i ]
A[i , j ]  ( x kpost , u k 1 ,0)
x[ j ]
f [i ]
W[ i , j ]  ( x kpost
1 , u k 1 ,0)
w[ j ]
h[ i ] ~
H [i , j ]  ( x k ,0)
x[ j ]
h[i ] ~
V[i , j ]  ( x k ,0)
v[ j ]

Now we define a new notation for the prediction error,

e xpri
k
 x k  x kpri (3.15)

and the measurement residual

e zpri
k
 z k  z kpri (3.16)

Remember that in practice one does not have access to xk in Equation (3.15), it is the

actual state vector, i.e. the quantity one is trying to estimate. On the other hand, one
does have access to in Equation (3.16), it is the actual measurement that one is using
to estimate. Using Equations (3.15) and (3.16) we can write governing equations for
an error process as Equation (3.17)

e~xk  A( x k 1  x kpost
1 )  k (3.17)

e~zk  He~xk   k (3.18)

where k and  k represent new independent random variables having zero mean and

covariance matrices WQW T and VRV T , with Q and R as in Equation (3.3)

respectively. Notice that the Equations (3.17) and (3.18) are linear, and that they
closely resemble the difference and measurement Equations (3.1) and (3.2) from the
discrete Kalman filter. This motivates us to use the actual measurement residual in
Equation (3.16) and a second (hypothetical) Kalman filter to estimate the prediction
error given by Equation (3.17). This estimate, call it, ekpost could then be used along
with Equation (3.15) to obtain the posterior state estimates for the original non-linear
process as

17
 xkpost  ~
xk  ekpost (3.19)

The random variables of (3.17) and (3.18) have approximately the following
probability distributions

p(e~xk )  N (0, E[e~xk e~xTk ])

p(k )  N (0,WQkW T )
p( k )  N (0,VRkV T )

Given these approximations and letting the predicted value of be zero, the Kalman
filter equation used to estimate is

ekpost  K k ezpri
k
(3.20)

By substituting (3.20) back into (3.19) and making use of (3.16) we see that we do not
actually need the second (hypothetical) Kalman filter

xkpost  xkpri  ekpost

(3.21)
 xkpri  K k ezpri
k

Equation (3.21) can now be used for the measurement update in the extended Kalman
filter, with ~
xk and ~z k coming from (3.13), and the Kalman gain K coming from (3.8)
with the appropriate substitution for the measurement error covariance.

An important feature of the EKF is that the Jacobian in the equation for the Kalman
gain serves to correctly propagate or “magnify” only the relevant component of the
measurement information. For example, if there is not a one-to-one mapping between
the measurement z k and the state via h , the Jacobian H affects the Kalman gain so

that it only magnifies the portion of the residual z k  h( xkpri ,0) that does affect the

state. Of course if over all measurements there is not a one-to-one mapping between
the measurement zk and the state via h, then as you might expect the filter will

quickly diverge. In this case the process is unobservable.

18
Need for UKF

In the EKF the state distribution is approximated by a GRV which is then propagated
analytically through the “first-order” linearization of the nonlinear system. These
approximations, however, can introduce large errors in the true posterior mean and
covariance of the transformed (Gaussian) random variable, which may lead to sub-
optimal performance and sometimes divergence of the filter. It is these “flaws” which
will be amended in the next section using the UKF.

3.4 UNSCENTED KALMAN FILTER-UKF

The Unscented Kalman Filter (UKF) addresses the approximation issues of the EKF.
The state distribution is again represented by a GRV, but is now specified using a
minimal set of carefully chosen sample points. These sample points completely
capture the true mean and covariance of the GRV, and when propagated through the
true non-linear system, captures the posterior mean and covariance accurately to the
3rd order (Taylor series expansion) for any nonlinearity. To elaborate on this, we start
by first explaining the unscented transformation. The Unscented transformation
calculates the statistics of a random variable which undergoes a nonlinear
transformation. Consider propagating a random variable ‘x’ of dimension L through a
nonlinear function, z  h(x) .To calculate the statistics of y, we introduce a matrix 
of length 2L+1 sigma vectors  i

where
~
0  X
~
  X  ( (L   )P ) , i  1,...., L
i x i
~
  X  ( (L   )P ) , i  L  1,...., 2 L
i x i
( m) 
W 
0 L
(3.22)
(c ) 
W   (1   2   )
0 L
( m) (c ) 1
W W  , i  1,....,2L
i 0 2( L   )
   2 (L  K )  L
   2 (L  K )  L
19
where  and K are scaling parameters.

 is usually set to a small positive value ( 103 to 1)

K is usually set to 0 and for Gaussian distributions,  =2

The above calculated sigma vectors around X are propagated through the nonlinear
function i.e. then we get

Zi  h( i ), i  0,....., 2L

and the mean and covariance for y are approximated us ing a weighted sample mean
and covariance of the posterior sigma points,

pri 2 L (m)
Z   W Z (3.23)
k i i
i0
2L
Pz   Wi c [ Z i  Z pri ][ Z i Z pri ]T
i 0

The UKF is a straight forward extension of the unscented Transformation to recursive

estimation as shown in Equation (3.24). Note that no explicit calculation of Jacobians
is necessary to implement this algorithm.

post pri pri

x X  K (Z  Z ) (3.24)
k k k k k

20
CHAPTER 4

APPLICATION OF KALMAN FILTER TECHNIQUES IN

POWER SYSTEM STATE ESTIMATION

In this chapter we discuss how these Kalman filter algorithms have been used for
power system state estimation through following sections.

4.1 POWER SYSTEM STATIC STATES ESTIMATION

USING DKF AND EKF

4.1.1 Working Methodology

A Prerequisites for applying the working methodology

 Choose the power system network for which you want to find states. In this
project we chosen IEEE-14 bus test system is considered [9] which consists of
5 generator buses and 11 load buses as shown in Figure 4.1. Bus 1 is
considered as slack bus.

Figure 4.1 IEEE 5-generaor, 14-bus network

21
  Input data
a) Bus data
i) Phasor voltage of buses i.e. Voltage and its angle at all buses.
ii) Injected Real and Reactive power at all buses

iii) Load Real and Reactive power at all buses

b) Line data

i) Line reactance and resistances & shunt admittance values

c) Measurement data
i) Real and Reactive power Injections
ii) Real and Reactive power flows
 Decide the output i.e. which states you want to estimate. Here we are
considering static states i.e. Voltage and its angle at all buses.

4.1.2 Static States Estimation Using DKF

We have discrete time controlled process x k with measurement z k that is governed

by the linear stochastic difference equation

xk  Ax k 1  Bu k 1  wk 1 (4.1)

z k  Hx k  vk

Static states include both voltage and its angle so and so the actual power injection
equation is

Vi V ( g
jN Li
i ij cos( ij )  bij sin( ij )) (4.2)

For applying DKF algorithm we need to approximate above equation by assuming  ij

small,loss less lines and voltage magnitude of 1 pu at all buses. Then it becomes
linear which results into

b 
jN Li
ij ij (4.3)

22
where

 ij is the voltage angle between the buses i, j

bij is the susceptance between the buses i, j

N Li Indicates number of buses connected to i th bus

As there is no control input B=0 and measurements considered are also constant as we
are not considering any fault case, A =I i.e. Identity matrix. Similarly consider noise
variables as mentioned in earlier chapter.

4.1.3 Algorithm of DKF

Time update equations

x kpri  x kpost
1
(4.4)
Pkpri  IPk 1 I T  Q

Measurement update equations

K k  Pkpri H T ( HPkpri H T  R) 1
x kpost  x kpri  K ( z k  Hx kpri ) (4.5)
Pkpost  ( I  K k H ) Pkpri

The time update equations are for predicting the state variables. The measurement
update equations are for correcting or filtering the state variables.

4.1.4 Static States Estimation Using EKF

We have discrete time controlled process that is governed by the non-linear stochastic
difference equation

x k  f ( x k 1 , u k 1 , wk 1 )
(4.6)
z k  h( x k , v k )

As EKF can be applied to non-linear process here there is no need to approximate

power injection equation i.e.

Vi V ( g
jN Li
i ij cos( ij )  bij sin( ij )) (4.7)

23
So using EKF we can estimate both the state variable without approximating the the
above non linear equation which is the main advantage of EKF over DKF.

As there is no control input B=0 and measurements considered are also

constant as we are not considering any fault case, A =I i.e. Identity matrix. Similarly
consider noise variables as mentioned in earlier chapter. Then apply DKF algorithm
by considering table as follows

4.1.5 Algorithm of EKF

Time update equations

pri post
x x
k k 1
(4.8)
pri
P  IP I T  WQW T
k k 1

Measurement update equations

K  P H T ( HP H T  VRV T )  1
k k
post pri pri
X X  K ( z  h( X ,0)) (4.9)
k k k k
post pri
P  ( I  KH ) P
k k

The time update Equations are for predicting the state variables. The measurement
update equations are for correcting or filtering the state variables

4.2 POWER SYSTEM DYNAMIC STATES ESTIMATION

USING EKF AND UKF

4.2.1 Working Methodology

Prerequisites for applying the working methodology

 Choose the power system network for which you want to find states. In this
project we chosen IEEE-14 bus test system is considered [9] which consists of
5 generator buses and 11 load buses as shown in Figure 4.1. Bus 1 is
considered as slack bus.

24
 Figure 4.1 IEEE 5- Generator, 14-Bus network

 Input data
a) Bus data
i) Phasor voltage of buses i.e. Voltage and its angle at all buses.
ii) Injected Real and Reactive power at all buses

iii) Load Real and Reactive power at all buses

b) Line data, Generator dynamic data

i) Line reactance and resistances & shunt admittance values

ii) H, M & xd values of generators

c) Measurement data

i) Real and Reactive power Injections at all buses

ii) Voltage and its angle at all buses
 Decide the output i.e. which states you want to estimate. Here we are
considering dynamic states i.e. relative angular speed and rotor angle of
generators..
Swing Equation
Under normal operating conditions, the relative position of the rotor axis and the
resultant magnetic field axis is fixed. The angle between the two is known as the
power angle or torque angle. During any disturbance, rotor will decelerate or

25
accelerate with respect to the synchronously rotating air gap MMF, a relative
motion begins. The equation describing the relative motion is known as the swing
equation. Consider a synchronous generator with electromagnetic torque Te running

at synchronous speed m . During the normal operation, the mechanical torque

Tm  Te . Any disturbance will result in accelerating/decelerating torque

Ta  Tm  Te ( Ta >0 if accelerating,

Ta <0 if decelerating).
By the law of rotation
d 2 m
J  Ta (4.10)
dt 2
 m  m t   m (4.11)

where
J is the combined moment of inertia of prime mover and generator

 m is the angular displacement of rotor w.r.t. stationary reference frame in rad

m is the synchronous speed in rad/s

d m d
 m 
dt dt
(4.12)
d m d m
2 2

dt 2 dt 2
From (4.10)
d 2 m
J  Ta
dt 2
(4.13)
d 2 m
Jm  Tam  Pa  Pm  Pe
dt 2

Swing equation in terms of inertial constant M

d 2 m
M  Pm  Pe (4.14)
dt 2
where M  J m

26
 Relation between electrical power angle,  e and mechanical power angle,  m is

given by Equation (4.15).

p
e  m (4.15)
2
and so
p
e  m (4.16)
2
Now Swing equation in terms of electrical power angle  e

2 d 2 e
M  Pa
p dt 2
 2   m2 J  d 2  e
2  
 2  dt 2  Pm  Pe
 P m  
2  1 J m  d  e Pm  Pe
2 2
  
we  2 S  dt 2 S

2 H d 2 e
 Pm  Pe (4.17)
 e dt 2

where
1 J m2
H
2 S
And is known as Inertia constant
Now, including damping power the swing equation becomes
2 H d 2 e
 Pm  Pe  Pd (4.18)
 e dt 2
And Pd is known as damping power

d 1
 ( Pm  Pe  Pd )
dt M
(4.19)
d

dt
where
 is the relative angular speed in rad/s
 is the rotor angle w.r.t. synchronously rotating frame in rad

27
 P is the Mechanical input power in pu which is constant
m
P is the Electrical output power in pu
e

Pe  E g I g
*

~ (4.20)
I g  YE g

P is the Damping power in pu

d
d
Pd  D (4.21)
dt

4.2.2 Algorithm of working methodology

1) Consider the entire power system network , In that apply nodal analysis .Then
generalise equations as
I Y V (4.22)
n bus n
where Vn is an (ng  nb )  1 column vector which includes generator internal

voltages and voltages at all buses. n g and nb are number of generators and number of

buses respectively. From the generalised Equation (4.22) find the matrix Ybus which is
of size (ng  nb )  (ng  nb ) . Here we need to find Y for both cases of power
bus
system i. e during normal operation and after fault condition.
1 1
I g1  Eg1  V1
jxd1 jxd1
E g 1  V1
I g1 
jxd1
E g 2  V2
Ig2 
jxd 2 (4.23)
E g n  Vn
I gn 
jxd n
E g1  V1
 (V1  V2 ) y12  (V1  V5 ) y15
jxd1
I g1  V1 ( y12  y15 )  V2 y12  V5 y15

28
 E g 2  V2
 I L  V2 ( y 21  y 23  y 24  y 25 )  V1 y 21  V3 y 23  V4 y 24  V5 y 25 (4.24)
jxd 2

We have

I g n  YbusV g n (4.25)

where

 I g1 
 
I g2 
I 
 g3 
I g 
 4
 I g5 
  I 
Ign  0    g
0  0
 
. 
. 
 
. 
0 
 
  191

 E1 
E 
 2
 E3 
 
E4 
 E5 
 
V  E 
Vgn   1    g
V
 2  Vb 
V3 
 
. 
. 
 
. 
V 
 14  191
Ygg Ygb 
Ybus   
Ybg Ybb 

29
  1 
 jx 0 0 0 0 
 d1 
 0 1
0 0 0 
 jxd 2 
 
1
Ygg  0 0 0 0 
 jxd3 
 1 
 0 0 0 0 
 jxd 4 
 1 
 0 0 0 0 
 jxd 5 

 1 
 jxd1
0 0 0 0 0 0 0 0 0 0 0 0 0
 
 0 
1
0 0 0 0 0 0 0 0 0 0 0 0
 jxd 2 
 
1
Ygb   0 0  0 0 0 0 0 0 0 0 0 0 0
 jxd3 
 1 
 0 0 0 0 0  0 0 0 0 0 0 0 0
 jxd 4 
 1 
 0 0 0 0 0 0 0  0 0 0 0 0 0
 jxd5 

 y12  y15  y12 0 0  y15 0 0 0 0 0 0 0 0 0 

 y y21  y23  y24  y25  y23  y24  y25 0 0 0 0 0 0 0 0 0 
 12 
 0  y12 . . . . . . . . . . 0 0 
 
 0  y12 . . . . . . . . . . 0 0 
  y15  y12 . . . . . . . . . . 0 0 
 
 0 0 . . . . . . . . . .  y613 0 
 0 0 . . . . . . . . . . 0 0 
Ybb     YLG
 0 0 . . . . . . . . . . 0 0 
 0 0 . . . . . . . . . . 0  y914 
 
 0 0 . . . . . . . . . . 0 0 
 
 0 0 . . . . . . . . . . 0 0 
 0 0 . . . . . . . . . .  y1213 0 
 
 0 0 . . . . . . . . . . y613  y1213  y1314  Y1314 
 0 0 0 0  y914  y1314 y914  y1314 
 0 0 0 0 0 0 0

Ybg  Ygb

y1314 is the admittance between the buses 13 and 14. Similarly other buses
admittances can also be referred.

30
Now we have

S L  PL  jQL
S L  V L YL
* 2

PL  jQL
YL  2
VL
I g n  YbusV g n
 I g  Ygg Ygb   E g 
 0   Y  
   bg Ybb   Vb 
I g  Ygg E g  YgbVb
0  Ybg E g  YbbVb
Ybb
Eg   Vb
Ybg
I g  (Ygg  YgbYbb1Ybg ) E g
~
I g  YE g (4.26)
~
Y  (Ygg  YgbYbb1Ybg )

2) We need to convert the differential equations of swing equation into difference

equations for finding states at all instants of time during our observation,
which are useful for all digital applications. Here choosing Runge-Kutta 2nd
order method for converting differential equations into difference equations.

Runge-Kutta 2nd order method

dy
 f ( x, y ), y (0)  y 0
dx
y i 1  y i  (a1 k1  a 2 k 2 )h
k1  f ( x i , y i )
k 2  f ( xi  lh, y i  mk1 h) (4.27)
a1  a 2  1
1
a2l 
2
1
a2 m 
2

Since we have 3 equations and 4 unknowns, we can assume the value of one of the
unknowns. The other three will then be determined. Generally the value of a2 is

31
chosen to evaluate the other three constants. The three values generally used for a 2
1 2
are , 1 and and are known as Heun’s Method, the midpoint method and
2 3
Ralston’s method, respectively.

Heun’s method

1
If we consider a 2 = and then solving equation we’ll get
2

1
a1 
2
l 1
m 1

So the required solution becomes

1 1
yi 1  yi  ( k1  k 2 )h (4.28)
2 2

where

k1  f ( xi , yi )
k 2  f ( xi  h, yi  k1h)
h  xi 1  xi

First find the true states of the power system network by considering the
measurements without noise so that we can compare these with estimated states of
power system network by considering measurements with noise. Then find the
estimated states of power system network by using the EKF and UKF algorithms as
described in following sections.

4.2.3 Dynamic States Estimation Using EKF

We have discrete time controlled process that is governed by the non-linear stochastic
difference equation

x  f (x ,u ,w )
k k 1 k 1 k 1
(4.29)
z  h( x , v )
k k k

32
Here the measurements considered are not constant fault case, so

f (4.30)
A  |
k x X post
k 1

where

f is the state variable equation

Similarly consider noise variables as mentioned in earlier chapter. Then apply EKF
algorithm by considering table as follows

4.2.4 Algorithm of EKF

Time update equations

pri pri
x  f (x ,u ,0)
k k 1 k 1 (4.31)
pri
P  A P AT  W Q W T
k k k 1 k k k 1 k

Measurement update equations

K  P H T (H P H T  V R V T )  1
k k k k k k k k k
post pri pri
x x  K ( z  h( X ,0)) (4.32)
k k k k k
post pri
P  (I  K H )P
k k k k

4.2.5 Dynamic States Estimation Unscented Kalman Filter

We have discrete time controlled process that is governed by the non-linear stochastic
difference equation

x  f (x ,u ,w )
k k 1 k 1 k 1
(4.33)
z  h( x , v )
k k k

where f is the state variable difference equation

Similarly consider other required variables as mentioned in earlier chapter. Then

apply UKF algorithm by considering table as follows

33
4.2.6 Algorithm of UKF

Time update Equations

pri 2 L ( m) x
x   W  i, k / k  1
k i
i0
pri 2 L (c ) pri pri T
P   W [ x X ][ x X ] W Q WT
k i i, k / k  1 k i, k / k  1 k k k 1 k
i0
pri 2 L ( m)
Z   W z
k i i, k / k  1
i0
(4.34)

Measurement update equations

2 L (c ) pri pri T
P   W [Z Z ][ Z Z ]  V RV T
z i i, k / k  1 k i, k / k  1 k k k
i0
2 L (c ) pri pri T
P   W [ Z ][ Z Z ]
xz i i, k / k  1 k i, k / k  1 k
i0
K  P P 1
xz z
post pri
x x  K ( Z  Z pri )
k k k k
post pri
P P  KP K T
k k z
(4.35)

34
CHAPTER 5

RESULTS AND DISCUSSION

5.1 POWER SYSTEM STATIC STATES ESTIMATION

In this study, we estimated static states i.e. the voltage and its angle at all buses of the
IEEE 5 generator, 14 bus systems by performing DKF and EKF algorithms through
MATLAB coding and the results have been shown in following figures and
table.Figure.5.1 shows the calculated active power flows values using estimated state
variables by DKF algorithm. Similarly Figure.5.2 shows calculated active power
flows values using estimated state variables by EKF algorithm. Table 5.1 gives the
calculated Average Error using DKF and EKF through estimates and measurements.

Calculate the estimated and measured square error by using Equation 5.1 and
5.2 respectively. The following table 5.1 gives the calculated square errors at one
sample point through estimates and measurements respectively using both DKF and
EKF respectively.

Estimated Square Error = ( x  x est ) 2 (5.1)

Measured Square Error = ( x  x meas ) 2 (5.2)

where

( x  xest )2 is the square of the difference between the true quantity and the

estimated quantity at an instant

( x  x meas ) 2 is the square of the difference between the true quantity and the

measured quantity at an instant

35
 Table 5.1 Estimated and Measured Square Error of DKF and EKF

Algorithm Estimated Measured

Square Error Square Error

DKF 7.9373 0.1306

EKF 0.0460 0.1009

2
True values
Active Power flows (pu)

1.5 Through estimates

Through measurements
1

0.5

-0.5

-1
0 2 4 6 8 10 12
K
Figure 5.1 Calculated Power flows using DKF Algorithm

2
True values
Active power flows (pu)

1.5 Through estimates

Through measurements
1

0.5

-0.5

-1
0 2 4 6 8 10 12
K
Figure 5.2 Calculated Power flows using EKF Algorithm

36
5.1.1 Observations

 From Figure 5.1 we can observe that the calculated pu active power flow
values using estimated states by DKF algorithm are not following the true pu
active power flow values. This is because of the approximation of the
nonlinear power injection equation to linear equation to perform DKF.
 From the Table 5.1 we can observe that the calculated estimated square error
value 7.9373 using DKF algorithm more than the measured square error
0.1306, which is not a favourable case. So we need to go for non linear filter
so that estimated square error will be less than measured square error.
 From Figure 5.2 we can observe that the calculated pu active power flow
values using estimated states by EKF algorithm are closely following the true
pu active power flow values. This is because EKF is a non linear filter and so
we can use the exact non linear equation for estimation without approximating
to linear equation as in the case of DKF.
 From the Table 5.1 we can observe that the calculated estimated square error
value 0.0460 using EKF algorithm is less than the measured square error
0.1009, which is a favourable case.

5.2 POWER SYSTEM DYNAMIC STATES ESTIMATION

As we are estimating dynamic states of the generators in IEEE 5 genrator,14 bus

system, considered two cases as configuration topology error and sudden load change
conditions to know how adaptive these two algorithms for those changes. We applied
those two conditions between two samples during our observation time and the results
obtained have been shown in below figures.

Case: 1) During Configuration topology Error

In power system network there may be a chance of configuration topology error

which occurs due to wrong status of circuit breakers. To apply that condition we
removed a line between the buses 4 and 5 is removed for a time duration of 5
milliseconds between the instants 1601 to 1602.The both algorithms UKF and EKF
have been tested for this case. We can observe the changes in the true and estimated
state variables during configuration topology error from the figures shown

37
below.Figure.5.3 shows the variation of actual and estimated values of rotor angle of
generator 2 whereas the fig.5.4 shows the true and estimated values of rotor speed of
generator 2.Similarly Figure.5.5 and Figure.5.6 shows the actual and estimated values
of rotor angle and rotor speed of generator 3 respectively during configuration
topology error.
Case: 2) During Load Change
The above algorithms are tested on an IEEE 14 bus power system network by
applying sudden load change as another Case. The load has been changed from 94.2
MW to 294.2 MW at bus 3 and it is applied for the time duration of 5 milliseconds
between the instants 1601 to 1602. We can observe the changes in the true and
estimated state variables during load change from the figures shown below. Figure.5.7
shows the variation of true and estimated values of rotor angle of generator 2 whereas
the Figure.5.8 shows the actual and estimated values of rotor speed of generator
2.Similarly Figure.5.9 and Figure.5.10 shows the actual and estimated values of rotor
angle and rotor speed of generator 3 respectively during load change.
Calculate the average estimated error by using equation (10). Table 5.2 shows
the average estimated error values using UKF and EKF for both the cases of load
change and configuration topology error.

1
Average Estimated Error =
ni  ns
| X  X post
| (5.3)

where

ni is the Total number of Instants

ns is the Total number of States

| X  X post
| is the sum of absolute values of difference between the true

states and the estimated states at all instants

38
 Table 5.2 Average Estimated Error of EKF and UKF

Algorithm Average Estimated Average

Error for configuration Estimated Error
change for load change

EKF 0.0403 0.0320

UKF 0.0355 0.0317

30 25
2
25 20 ekfest
Rotor angle, 2(rad)

2
20 15 ukfest
2

15 10

10 5

5 0
8 8.005 8.01 8.015
0
6 8 10 12 14 16 18 20
Time (s)
Figure 5.3 Variation of Rotor Angle of Generator 2 using EKF and UKF Algorithms
during Configuration Topology Error

39
 10 2
10
Rotor Speed, 2(rad/s)

ekfest
2
5 ukfest
2
5
0

0 7.995 8 8.005 8.01 8.015

6 8 10 12 14 16 18 20
Time (s)
Figure 5.4 Variation of Angular Speed of Generator 2 using EKF and UKF
Algorithms during Configuration Topology Error

12
3
10 10
ekfest
Rotor angle, 3 (rad)

3
8
ukfest
3
6 5
4
2
0
0 7.98 7.99 8 8.01 8.02
-2
6 8 10 12 14 16 18 20
Time (s)
Figure 5.5 Variation of Rotor Angle of Generator 3 using EKF and UKF Algorithms
during Configuration Topology Error

40
 5 4
3
Angular speed, 3(rad/s)
4 3 ekfest
3
3 2 ukfest
3

2 1

1 0

0 8 8.05 8.1 8.15

-1
6 8 10 12 14 16 18 20
Time (s)
Figure 5.6 Variation of Angular Speed of Generator 3 using EKF and UKF
Algorithms during Configuration Topology Error

1.5
2
1.5
Roto angle, 2 (rad)

ekfest
1 2
ukfest
2
1
0.5

0.5 0
8 8.005 8.01 8.015 8.02 8.025

0
6 8 10 12 14 16 18 20
Time (s)
Figure 5.7 Variation of Rotor Angle of Generator 2 using EKF and UKF Algorithms
during Sudden Load Change

41
 0.2
2
Angular Speed, 2(rad/s)

0 ekfest
2
0 -0.2 ukfest
2
-0.4

-0.5 -0.6
-0.8
8 8.01 8.02 8.03
-1
6 8 10 12 14 16 18 20
Time (s)

Figure 5.8 Variation of Relative Angular Speed of Generator 2 using EKF and UKF
Algorithm during Sudden Load Change

1 1 3
Rotor angle, 3(rad)

0.8 ekfest
3
ukfest
0.6 0.5 3

0.4
0.2
0
0 8 8.05 8.1
-0.2
6 8 10 12 14 16 18 20
Time (s)
Figure 5.9 Variation of Rotor Angle of Generator 3 using EKF and UKF Algorithm
during Sudden Load Change

42
 0 0 3
Rotor speed, 3(rad/s)

-0.2 ekfest
3

-0.4 ukfest
-0.5 3

-0.6
-0.8 -1
-1 8 8.02 8.04 8.06
-1.2
6 8 10 12 14 16 18 20
Time (s)
Figure 5.10 Variation of Relative Angular Speed of Generator 3 using EKF and UKF
Algorithm during Sudden Load Change

5.2.1 Observations

Case: 1) During Configuration Topology Error

 From Figure 5.3 and 5.4, we can observe the variation of rotor angle of
generator 2 i.e. and the variation of angular speed of generator 2 i.e. for a
duration of 20 seconds which includes the configuration topology error. The
figure shows clearly that the estimated state variables of generator 2 using
UKF is closely following the true calculated state variable than the estimated
state variables using EKF algorithm. We can also observe that during the line
removal/configuration topology error the speed of the generator 2 is increasing
from its synchronous speed.
 From Figure 5.5 and 5.6, we can observe the variation of rotor angle of
generator 3 i.e. and the variation of angular speed of generator 3 i.e. for a
duration of 20 seconds which includes the configuration topology error. The
figure shows clearly that the estimated state variables of generator 3 using
UKF is closely following the true calculated state variable than the estimated
state variables using EKF algorithm. We can also observe that during the line
removal/configuration topology error the speed of the generator 3 is increasing
from its synchronous speed.

43
  From the table 5.2, we can observe that calculated Average Estimated Error
during configuration change by UKF algorithm is 0.0355 which is less than
that of EKF algorithm i.e. 0.0403

Case: 2) During Sudden Load Change

 From Figure 5.7 and 5.8, we can observe the variation of rotor angle of
generator 2 i.e. and the variation of angular speed of generator 2 i.e. for a
duration of 20 seconds which includes the Sudden load change i.e. increase in
load at bus 3. The figure shows clearly that the estimated state variables of
generator 2 using UKF is closely following the true calculated state variable
than the estimated state variables using EKF algorithm. We can also observe
that during the load change, the speed of the generator 2 is decreasing from its
synchronous speed.
 From Figure 5.9 and 5.10, we can observe the variation of rotor angle of
generator 3 i.e. and the variation of angular speed of generator 3 i.e. for a
duration of 20 seconds which includes the Sudden load change i.e. increase in
load at bus 3. The figure shows clearly that the estimated state variables of
generator 3 using UKF is closely following the true calculated state variable
than the estimated state variables using EKF algorithm. We can also observe
that during the load change, the speed of the generator 3 is decreasing from its
synchronous speed.
 From the table 5.2, we can observe that calculated Average Estimated Error
for the case of Sudden load change by UKF algorithm is 0.0317, which is less
than that of EKF algorithm i.e. 0.0320

44
CHAPTER 6

CONCLUSIONS AND FUTURE SCOPE

6.1 CONCLUSIONS

1. From MATLAB results shown above for static states estimation, we can say that
for nonlinear functions EKF is far better than DKF even though if we use the
approximated linear function for DKF in case of power system application. The Error
along with the measurements has been reduced using EKF. With this we can conclude
that we can’t approximate the actual nonlinear power equations to find state estimates
using DKF which results in more erroneous values than measurement values.

2. From MATLAB results shown above for dynamic states estimation, we can say
that the estimated dynamic states i.e. rotor angle and rotor speed of the IEEE-14 bus
power system network using UKF and EKF algorithms are following the true values
of the dynamic states of the network during normal and transient case like sudden
load change and configuration topology error. The calculated average estimated errors
from the table show that the error using UKF is comparatively less than the error
using EKF for both the cases, which means the UKF is more adaptive than EKF in
dynamic state estimation of power system network.

6.2 FUTURE SCOPE

In this project SCADA measurement data is used which is less accurate than Phasor
measurement units (PMUs) data. A phasor measurement unit (PMU) or
synchrophasor is a device which measures the electrical waves on an electricity grid,
using a common time source for synchronization. Time synchronization allows
synchronized real-time measurements of multiple remote measurement points on the
grid. So in future for more accuracy we can implement these algorithms using PMU
measurement data.

45
PUBLICATIONS

[1] Manojkumar Rampelli and Debashisha Jena, “Advantage of Unscented Kalman

Filter over Extended Kalman Filter in Dynamic State Estimation of Power System
Network,” Michael Faraday IET International Summit: MFIIS-2015, September 12 –
13, 2015, Kolkata, India.[Accepted]

46
REFERENCES

[1] F. C. Schweppe and J. Wildes, "Power system static-state estimation,pt. I: exact

model," this issue, pp. 120-125.

[2] F. C. Schweppe and D. Rom, "Power system static-state estimation, pt.II:

approximate model," this issue, pp. 125-130.

[3] F. C. Schweppe, "Power system static-state estimation, pt.III: implementation,"

this issue, pp. 130-135.

[4] D. M. Falcao, P. A. Cooke, and A. Brameller, “Power system tracking state

estimation and bad data-processing,” IEEE Trans. Power Appl. Syst., vol. PAS-101,
no. 2, pp. 325–333, Feb. 1982..

[5] M. B.DoCoutto Filho and J. C. Stacchini de Souza, “Forecasting-aided state

estimation—Part I: Panorama,” IEEE Trans. Power Syst., vol. 24, no. 4, pp. 1667–
1677, Nov. 2009.

[6] M. B.DoCoutto Filho and J. C. Stacchini de Souza, “Forecasting-aided state

estimation—Part II: Panorama,” IEEE Trans. Power Syst., vol. 24, no. 4, pp. 1667–
1677, Nov. 2009.

[7] A. Jain and N. R. Shiva kumar, “Power system tracking and dynamic state
estimation,” in 2009. PSCE ’09. IEEE/PES, Power Systems Conference and
Exposition, 2009, pp. 1–8.

[8] D. M. Falcão, P. A. Cooke, and A. Brameller, “Power system tracking state

estimation and bad data processing,” IEEE Trans. Power App.Syst., vol. PAS-101, pp.
325–333, Feb. 1982.

[9] A. M. Leite da Silva, M. B. Do Coutto Filho, and J. F. de Queiroz, “State

forecasting in electric power systems,”Proc. Inst. Elect. Eng. C,vol. 130, no. 5, pp.
237–244, Sep. 1983.

47
[10] A. M. Leite da Silva, M. B. Do Coutto Filho, and J. M. C. Cantera, “An efficient
dynamic state estimation including bad data processing,”IEEE Trans. Power Syst.,
vol. 2, no. 4, pp. 1050–1058, Nov. 1987.

[11] M. B. Do Coutto Filho, A. M. Leite da Silva, J. M. C. Cantera, and R. A. da

Silva, “Information debugging for real-time power systems monitoring,”Proc. Inst.
Elect. Eng. C, vol. 136, no. 3, pp. 145–152, May 1989.

[12] K. R. Shih and S. J. Huang, “Application of a robust algorithm for dynamic state
estimation of a power system,”IEEE Trans. Power Syst., vol. 17, no. 1, pp. 141–147,
Feb. 2002.

[13] S. J. Huang and K. R. Shih, “Dynamic state estimation scheme including

nonlinear measurement function considerations,”Proc. Inst.Elect. Eng. C, vol. 149, no.
6, pp. 673–678, Nov. 2002.

[14] J. M. Lin, S. J. Huang, and K. R. Shih, “Application of sliding surface-enhanced

control for dynamic state estimation of a power system,”IEEE Trans. Power Syst.,
vol. 18, no. 2, pp. 570–577, May 2003.

[15] J. C. S. Souza, A. M. Leite da Silva, and A. P. Alves da Silva, “Information

debugging in forecasting-aided state estimation using a pattern analysis approach,” in
Proc. 12th Power Syst. Comput. Conf. (PSCC), Dresden, Germany, Aug. 1996, vol.
II, pp. 1214–1220.

[16] J. C. S. Souza, A. M. Leite da Silva, and A. P. Alves da Silva, “Data debugging

for real-time power system monitoring based on pattern analysis, ”IEEE Trans. Power
Syst., vol. 11, no. 3, pp. 1592–1599, Aug.1996.

[17] J. C. S. Souza, A. M. Leite da Silva, and A. P. Alves da Silva, “Data

visualization and identification of anomalies in power system state estimation using
artificial neural networks, ”Proc. Inst. Elect. Eng. C, vol.144, no. 5, pp. 445–455, Sep.
1998.

[18] J. C. S. Souza, A. M. Leite da Silva, and A. P. Alves da Silva, “Online topology

determination and bad data suppression in power system operation using artificial
neural networks, ”IEEE Trans. Power Syst., vol.13, no. 3, pp. 796–803, Aug. 1998.

48
[19] M. B.DoCoutto Filho and J. C. Stacchini de Souza, “Forecasting-aided state
estimation—Part I: Panorama,” IEEE Trans. Power Syst., vol. 24, no. 4, pp. 1667–
1677, Nov. 2009.

[20] B. Sadecky and P. Neuman, “State estimation and short term load pre- diction as
components of the power system security assessment”,” in Proc. 11th Power Syst.
Computation Conf. (PSCC), Avignon, France, Aug. 1993, vol. II, pp. 705–712.

[21] A. P. Alves da Silva, A. M. Leite da Silva, J. C. Stacchini de Souza, and M. B.

Do Coutto Filho, “State forecasting based on artificial neural networks”,” in Proc.
11thPower Syst. Computation Conf. (PSCC),Avi- gnon, France, Aug. 1993, vol. I, pp.
461–468.

[22] R. J. Hyndman and A. B. Koehler, “Another look at measures of fore- cast

accuracy,” Int. J. Forecast., vol. 22, no. 4, pp. 679–688, Oct.-Dec. 2006.

[23] R. E. Kalman, A new approach to linear filtering and prediction problems,

Trans. ASME, J. Basic Engrg., ser. D, vol. 82, pp. 35-45, 1960.

[24] E. A. Blood, B. Krogh, and M. Ilic, “Electric power system static state estimation
through Kalman filtering and load forecasting,” in 2008 IEEE Power and Energy
Society General Meeting - Conversion and Delivery of Electrical Energy in the 21st
Century, 2008, pp. 1–6.

[25] H.H.Goh, C.W.Tai, Q.S.Chua, S.W.Lee, B.C.Kok, K.C.Goh and K.T.K.Teo,

“Dynamic Estimation of Power System Stability in Different Kalman Filter
implementations,” Proceedings of the 2014 IEEE NW Russia Young Researchers in
Electrical and Electronic Engineering Conference.

[26] Yun Yang, Wei Hu and Yong Min, “Projected unscented Kalman filter for
dynamic state estimation and bad data detection in power system, “12th IET
International Conference on Developments in Power System Protection (DPSP 2014).

[27] A. J. Wood, B. F. Wollenberg, “Power Generation, Operation, and Control”,

New York, John Wiley & Sons, 1996.

[28] Welch Greg, Bishop Gary, “An Introduction to Kalman Filter”, UNC-Chapel
Hill, TR 95-041, July 24, 2006

49
APPENDIX I

WEIGHTED LEAST SQUARES ESTIMATION

7.1 MAXIMUM LIKELIHOOD CONCEPTS

The principle of maximum likelihood estimation is illustrated by using a simple DC

circuit example as shown in Figure 6.1. In this example, we wish to estimate the value
of the current source which is connected as shown in Figure 6.1.Supplose if we
connect two voltmetes with an error having a known standard deviations, then those

Figure 7.1 DC Circuit Example Weighted Least Squares Estimation

meters connected across resistors r1 and r2 gives readings of z1meas and z2meas
respectively

z1meas  z1true  1
(6.1)
z2meas  z2true  2

where

The errors 1 and 2 represented as independent zero mean, normally distributed

random variables with probability density functions

50
  12 
 
1  2 2 
PDF (1 )  e 1 
 1 2
(6.2)
  22 
 
1  2 2 
PDF (2 )  e  2 

 2 2

We need to consider the product of the probability of obtaining z1meas and the

probability of obtaining z2meas for the probability of obtaining z1meas and z2meas . So
consider that product as

PDF ( z1meas )  PDF ( z2meas ) (6.3)

and then maximize that function, we will get

 ( z meas  xr1 ) 2 ( z2meas  xr2 ) 2 

max  Ln ( 1 2 )  1  Ln ( 2 )  
2 1 2 2
2 2 2
x
 
(6.4)
 ( z meas  xr1 ) 2 ( z2meas  xr2 ) 2 
min  1  
2 1 2 2
2 2
 
x

Then minimum sought is found by

d  ( z1meas  xr1 ) 2 ( z2meas  xr2 ) 2  2r1 ( z1meas  xr1 ) 2r1 ( z21

meas
 xr2 )
    0
2 1 2 2 2 1 2 2
2 2 2 2
dx  

z meas r   z meas r 
2 2
x est
 1 1 2 2 2 22 1 (6.5)
(r1 2 )  (r2 1 )

And if one of the voltmeter is of superior quality, its variance will be much smaller
than that of the other meter. Suppose  2 <<  1
2 2

z2meas
x est  (6.6)
r2

Thus, we see that the maximum likelihood method of estimating our unknown
parameter gives us a way to weight the measurements properly according to their
quality[27]. It should be obvious by now that we need not express our estimation
problem as a maximum of the product of probability density functions. In these
equations, we see that the maximum likelihood estimate of our unknown parameter is

51
always expressed as that value of the parameter that gives the minimum of the sum of
the squares of the difference between each measured value and the true value being
measured. If we are estimating a single parameter, x , using N m , measurements, we

would write the expression

Nm
( zimeas  f i ( x))
min J ( x)   (6.7)
x
i 1 i2

f i ( x1 , x2 ,........ xNs )  hi1 x1  hi 2 x2  .....  hiNs xNs

where

f i = function that is used to calculate the value being measured by the ith
measurement

 i 2 = variance for the ith measurement

J (x) = measurement residual

N m = number of independent measurements

zimeas = measured i th quantity

Using matrix notation now, write

 f1 ( x) 
 f ( x) 
 2 
f ( x)   .   [ H ]x
 
 . 
 f N ( x)
 m 

 z1meas 
 meas 
 z2 
z meas  . 
 
 . 
 z meas 
 Nm 

We can write equation in a compact form as

52
 min J ( x)  [ z meas  f ( x)]T [ R 1 ][ z meas  f ( x)] (6.8)
x

where

 12 . . . . 
 
 . 2
2
. . . 
[ R]   . . . . . 
 
 . . . .. . 
 . . . .  N m 
2


Finding the gradient of J (x) and make it equals to zero to find the minimum of J (x)

J ( x)  2[ H ]T [ R1 ]z meas  2[ H ]T [ R1 ][ H ]x (6.9)

Then J ( x)  0 , gives

xest  [[ H ]T [ R1 ][ H ]T ]1[ H ]T [ R1 ]z meas (6.10)

The advantages of using a state estimation algorithm in that, even with

measurement errors, the estimation algorithm calculates quantities that are the “best”
possible estimates of the true bus voltages and generator, load, and transmission line
MW and MVAR values. There are, however, other advantages to using a state
estimation algorithm. First, is the ability of the state estimator to detect and identify
bad measurements, and, second, is the ability to estimate quantities that are not
measured and telemetered.

53
APPENDIX II

IEEE 5 GENERATOR,14 BUS DATA

8.1 LINE DATA

Table 8.1 IEEE 5 generator,14 bus network line data

From To R X B/2
Bus Bus (pu) (pu) (pu)

1 2 0.01938 0.05917 0.0264

1 5 0.05403 0.22304 0.0246
2 3 0.04699 0.19797 0.0219
2 4 0.05811 0.17632 0.0170
2 5 0.05695 0.17388 0.0173
3 4 0.06701 0.17103 0.0064
4 5 0.01335 0.04211 0.0
4 7 0.0 0.20912 0.0
4 9 0.0 0.55618 0.0
5 6 0.0 0.25202 0.0
6 11 0.09498 0.19890 0.0
6 12 0.12291 0.25581 0.0
6 13 0.06615 0.13027 0.0
7 8 0.0 0.17615 0.0
7 9 0.0 0.11001 0.0
9 10 0.03181 0.08450 0.0
9 14 0.12711 0.27038 0.0
10 11 0.08205 0.19207 0.0
12 13 0.22092 0.19988 0.0
13 14 0.17093 0.34802 0.0

54
8.2 GENERATOR DYNAMIC DATA

Table 8.2 IEEE 5 generator,14 bus network generator dynamic data

Generator xd H D

(pu) (sec) (pu)

G1 0.2995 5.148 2
G2 0.185 6.54 2
G3 0.185 6.54 2
G4 0.232 5.06 2
G5 0.232 5.06 2

8.3 BUS DATA

Table 8.3 IEEE 5 generator,14 bus network bus data

Bus Type Vb  PGi QGi PLi Q Li

No. (pu) (MW)
(pu) (MW) (MW) (MW)

1 1 1.060 0 30 0 0 0
2 2 1.045 0 40 42.4 21.7 12.7
3 2 1.010 0 20 23.4 94.2 19.0
4 3 1.0 0 0 0 47.8 3.9
5 3 1.0 0 0 0 7.6 1.6
6 2 1.070 0 20 12.2 11.2 7.5
7 3 1.0 0 0 0 0 0
8 2 1.090 0 50 17.4 0 0
9 3 1.0 0 0 0 29.5 16.6
10 3 1.0 0 0 0 9.0 5.8
11 3 1.0 0 0 0 3.5 1.8
12 3 1.0 0 0 0 6.1 1.6
13 3 1.0 0 0 0 13.5 5.8
14 3 1.0 0 0 0 14.9 5.0

55
 56

View publication stats