Professional Documents
Culture Documents
net/publication/324782095
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.
Thesis
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
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.
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 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.
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 TABLES iv
LIST OF SYMBOLS v
1 INTRODUCTION 1
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
iii
LIST OF TABLES
iv
LIST OF SYMBOLS
wk : Process noise
vk : Measurement noise
zk : Vector of Measurements
e xpri
k
: Prediction Error
e zpri
k
: Measurement residual
H : Inertia Constant
M : Angular momentum
xd : Damping Reactance
v
m : Mechanical power angle
nb : Number of buses
Vb : Bus Voltages
vi
LIST OF ACRONYMS
vii
CHAPTER 1
INTRODUCTION
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.
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.
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
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.
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,
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
3.1 INTRODUCTION
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.
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.
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.
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)
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
p( w) N (0, Q)
(3.3)
p(v) N (0, R)
ekpri x k x kpri
(3.4)
ekpost x k x kpost
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).
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
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.
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)
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)
z k h( xk , vk ) (3.12)
where xk is the vector of state variables. wk and vk are process and measurement
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
xk ~
x A( xk 1 xˆk 1 ) Wwk 1
(3.14)
z ~z H ( x ~
k k x ) Vv
k k k
where
~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,
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 ]
e xpri
k
x k x kpri (3.15)
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)
where k and k represent new independent random variables having zero mean and
17
xkpost ~
xk ekpost (3.19)
The random variables of (3.17) and (3.18) have approximately the following
probability distributions
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
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
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.
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.
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
i0
2L
Pz Wi c [ Z i Z pri ][ Z i Z pri ]T
i 0
20
CHAPTER 4
In this chapter we discuss how these Kalman filter algorithms have been used for
power system state estimation through following sections.
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.
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
b) Line data
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
jN Li
i ij cos( ij ) bij sin( ij )) (4.2)
small,loss less lines and voltage magnitude of 1 pu at all buses. Then it becomes
linear which results into
b
jN Li
ij ij (4.3)
22
where
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.
x kpri x kpost
1
(4.4)
Pkpri IPk 1 I T Q
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.
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 )
Vi V ( g
jN 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.
pri post
x x
k k 1
(4.8)
pri
P IP I T WQW T
k k 1
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
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
c) Measurement data
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
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
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
Jm Tam Pa Pm Pe
dt 2
26
Relation between electrical power angle, e and mechanical power angle, m is
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
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
191
E1
E
2
E3
E4
E5
V E
Vgn 1 g
V
2 Vb
V3
.
.
.
V
14 191
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
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 YgbYbb1Ybg ) E g
~
I g YE g (4.26)
~
Y (Ygg YgbYbb1Ybg )
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
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.
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
Similarly consider noise variables as mentioned in earlier chapter. Then apply EKF
algorithm by considering table as follows
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
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
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
33
4.2.6 Algorithm of UKF
pri 2 L ( m) x
x W i, k / k 1
k i
i0
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
i0
pri 2 L ( m)
Z W z
k i i, k / k 1
i0
(4.34)
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
i0
2 L (c ) pri pri T
P W [ Z ][ Z Z ]
xz i i, k / k 1 k i, k / k 1 k
i0
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
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.
where
( x xest )2 is the square of the difference between the true quantity and the
( x x meas ) 2 is the square of the difference between the true quantity and the
35
Table 5.1 Estimated and Measured Square Error of DKF and EKF
2
True values
Active Power flows (pu)
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)
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.
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
| X X post
| is the sum of absolute values of difference between the true
38
Table 5.2 Average Estimated Error of EKF and UKF
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
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
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
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
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
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.
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
46
REFERENCES
[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.
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.
[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.
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.
[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.
[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).
[28] Welch Greg, Bishop Gary, “An Introduction to Kalman Filter”, UNC-Chapel
Hill, TR 95-041, July 24, 2006
49
APPENDIX I
meters connected across resistors r1 and r2 gives readings of z1meas and z2meas
respectively
z1meas z1true 1
(6.1)
z2meas z2true 2
where
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
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
Nm
( zimeas f i ( x))
min J ( x) (6.7)
x
i 1 i2
where
f i = function that is used to calculate the value being measured by the ith
measurement
f1 ( x)
f ( x)
2
f ( x) . [ H ]x
.
f N ( x)
m
z1meas
meas
z2
z meas .
.
z meas
Nm
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)
Then J ( x) 0 , gives
53
APPENDIX II
From To R X B/2
Bus Bus (pu) (pu) (pu)
54
8.2 GENERATOR DYNAMIC DATA
Generator xd H D
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
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