Power System States Estimations Using Kalman Filter

Murtaza Farsadi1, Farzad Mohammadzadeh Shahir2, Student Member, IEEE, and Ebrahim Babaei3, Senior
Member, IEEE
1
Faculty of Electrical and Computer Engineering, Urmia University, Urmia, Iran
[email protected]
2
Department of Electrical Engineering, Urmia Branch, Islamic Azad University, Urmia, Iran
[email protected]
3
Faculty of Electrical and Computer Engineering, University of Tabriz, Tabriz, Iran
[email protected]

Abstract dynamic state estimation are in the field of linear systems [11-
12]. These papers include linearizing the algebraic and
Power system state estimation plays an important role in the differential equations can be done by calculating the Jacobian
analysis of control and system stability and many other key matrix. Linear approximation leads to errors, which could be
functions such as economic analysis of load, optimal power effective in the whole of the time, especially for higher-power
flow (OPF), events analysis and gaining static and dynamic and complex systems [14]. The Jacobian matrix is performed for
algorithms with complex calculations. Using linear designs,
parameters of systems are based on state estimation.
recent papers have done research aimed at eliminating
Conventional state estimation due to low rates of update of
linearization and using unscented conversion and preventing the
SCADA system, cannot demonstrate the dynamic situations
calculation of Jacobian matrix at any repetition [3, 13-14].
(several seconds). In order to deal with the obstacles due to Power system state estimation plays an important role in the
linearization and calculation of Jacobian matrix, a non-linear analysis of control and system stability and many other key
dynamic state estimation without calculation of the Jacobian functions such as economic analysis of load, OPF, events
matrix is provided in this paper. Kalman filter is a analysis and gaining static and dynamic parameters of systems
mathematical method that aims to provide values closer to are based on state estimation [15-16]. Conventional state
actual values from measurements that include noise and estimation due to low rates of update of SCADA system, cannot
inaccuracy. The method matches the non-linear demonstrate the dynamic situations (several seconds). In order
characteristics of system as well as probability of measured to deal with the obstacles due to linearization and calculation of
values. After, Kalman filter state estimation method is Jacobian matrix, a non-linear dynamic state estimation without
presented in both linear and nonlinear modes and simulation calculation of the Jacobian matrix is provided in this paper. At
results are presented in MATLAB. the beginning, it is tried to simply explain state estimation using
[17] and then its application and its relationship with other
1. Introduction sectors in the power system are elaborated.
In the following, to better understand the paper a linear
The structure of the transmission system in the developed system, which in fact are the equations of motion is used and
world is faced with two major challenges: reliability and simulation results are presented in MATLAB. Kalman filter has
extensibility. New technologies to produce electric power by errors when the input data is not obtained from dynamics
wind and solar energy replace fossil fuels in power plants with simulation, so using DigSilent software dynamic state of system
synchronized machine [1-2]. With these changes due to is modeled simulation results in is provided at this software but
concentrated synchronized production the current power model it should be noted that this data is not used in MATLAB
in transmission system cannot be fixed. In recent decades for software.
many reasons there have been no effective investments in the
transmission system [3-4]. Technologies related to control and 2. State Estamition
computational layers play an important role in avoiding the
blackouts of the transmission system. But the majority of First, we need to be discussing state estimation to some
monitoring and control instruments of energy management extent and provide an example using reference [17] to properly
systems (EMSs) are based on steady-state model of the system. understand the concept. In this reference, the concept has been
Therefore, it cannot be properly show dynamic performance of simply stated for static state. In the eighteenth century for the
the system [5-6]. This limit is due to dependency of EMS to first time, scientists observed wonderful arrangement in
very slow update rate of SCADA systems. With the advent of measuring errors. They found that observed samples or
phasor measurement unit (PMU) based on wide area distributions are approximated by continuous distribution as it
measurement systems (WAMSs) new techniques for dynamic was called normal curve of errors, and attributed them to the
security assessment (DSA) are developed to assess the security laws of chance. An operator must have access to the right
of system that is an effective method for studying the dynamics information from the system, to detect errors in the system based
of the system compared to conventional EMSs [7-10]. Exact on that information and based on this information, he can decide
knowledge of dynamic state of system is through dynamic state on overvoltage, overload etc. The task of state estimation is to
estimation (DSE) and it can be said that state estimation is the provide right information for operator at different conditions,
core of these techniques [11-13]. Most studies related to what made the state estimation successful is the advantage of

808
uthorized licensed use limited to: Indian Institute Of Technology (Banaras Hindu University) Varanasi. Downloaded on November 03,2024 at 17:12:07 UTC from IEEE Xplore. Restrictions apply
 probability rules; as mentioned above, the measurement errors these measurements has the risk of bad measurement; we will
are normally distributed and when a non-measurement error begin with the largest value and put it aside. Again state
exist in calculations it can be detected with this method and its estimation is performed for the rest of the data and J(x) will be
effect can be reduced or eliminated. Measuring converters in the specified; on the basis of the new threshold, if it was less than
power system, like any other device are prone to error, if these threshold, a bad measurement is achieved; otherwise, the
errors are small they may not be detected and measured value operation must be repeated for other candidates. It is necessary
can incorrectly enter the calculation and even the output may be to mention here that some of the measurements are with very
useless; for example, the device may be installed in reverse small errors. Therefore, it is not possible to detect them, but
order and may provide a negative value. In some cases, parts of what is important is that the operator knows that there is no big
telecommunication system may be disconnected and the error in the data.
operator may not be available; these were the reasons for
development of state estimation at the past.
Networks used in [17] are provided in Figure 1. With state
estimation, using the algorithm in Figure 2 values of Table I is
achieved. Random error is added to the measured values to
obtain the measurement with the random error. According to
Table I, it is possible to achieve efficiency of state estimation in
the power system. For example, measurement of M23 shows the
active power to be 8.6MW, but the actual value, 2.9MW, the value
proposed by estimator is equal to 3MW. Despite the measurement
error, the estimation algorithm calculated some values, which is
much closer to the actual values. There are other advantages for
state estimation algorithm including, a) The ability of state
estimation for detection bad measurements, b) The ability for
estimation of the values of which are not measured or
communicated. The following a brief explanation of reference
[17] is given.

Fig. 2. Algorithm used to state estimation of Figure 1

2.2. Estimation of the Values of Which are not

Measured or Communicated
Another important feature of state estimation is estimation of
the values of which cannot be transmitted. It may be due to
problems in telecommunications, as well as non-installation of
measuring instruments and transfer of it to the control center. If
in Figure 1, data from all buses of 3, 4, 5 and 6 is not
transmitted to the control center, still we can use the state
estimation algorithm, the values of voltage and phase of other
buses can be estimated and then generation, currents and loads
can be achieved; the results are shown in Table II. As can be
seen in Table II, estimated values are not very close in
Fig. 1. The circuits used in reference [17] to better understand comparison with when the full data is transmitted.
the state estimation in power system
2.3. Application of State Estimation in Power System
2.1. State Estimation for Detection of Bad Measuring the analog output of the generators should go
Measurements directly to the AGC program and other data before it is used by
The ability for detection of bad measurement in the power other programs must be processed in state estimation program.
system is valuable for operators. The converters may be wired To run the state estimation, first network topology (connection
incorrectly or may have a wrong function. Identifying these of line 5 to bus and load, on/off state of switches) must be
errors using statistical theory is very simple but lengthy. By specified, then the electric model of power transmission system
checking the value of J(x) you can understand the existence of a will be sent state estimation program. The output of state
bad measurement in the system, so that a threshold is defined estimation include voltage and phase all buses, current MW and
based on the probability, so if the Jacobian is less than this value MVAr in lines, bus loads and product of generators. These
(tj), it is supposed that there is no bad measurement in the values and electrical model are a basic for economic distribution
system, but if it is larger than this threshold, there is a bad of load, investigation of events and correction of production.
measurement in this system. Since, we have the complete electric model of transmission lines
But the question that arises here is that what the reason for so the penalty factor can be calculated as well.
bad measurement is in the system to be corrected. If the
difference between predicted and measured is shown with yi, it 3. Linear Kalman Filter
has a normal distribution with mean of zero and standard
deviation of yi, then it will be normalized so it has a standard Kalman filter is a mathematical method that aims to provide
deviation of 1. If the normalized value is larger than 3 cases values closer to actual values from measurements that include

809
uthorized licensed use limited to: Indian Institute Of Technology (Banaras Hindu University) Varanasi. Downloaded on November 03,2024 at 17:12:07 UTC from IEEE Xplore. Restrictions apply
 noise and inaccuracy. From the standpoint of theory, Kalman Table 1. The values obtained from the state estimation of
filter is an algorithm to infer an accurate result in a linear figure 1
dynamic system and all variables have Gaussian distribution.
The first implementation of the Kalman filter returns to his visit
from NASA research center that he found that he can use this
method to estimate the Apollo route. Kalman filter of system
dynamics model that control inputs and measurement systems
comprise it provide better value for the estimation of the system
variables compared to the individual measurements. First, to be
better understand the Kalman filter in non-linear mode, an
understandable example is expressed to specify the concepts
used in non-linear mode. For example, the aim is to determine
the exact position of a vehicle, which there is a GPS in the
vehicle that can locate the device with an error of up to several
meters; the estimation of GPS is in the form of noise that
sometimes sudden jumps can be observed, but there is always
maximum error. The position of vehicle can through be obtained
taking integration of speed and direction over time. It means that
by gaining changes of momentum and changes of wheel the data
can be obtained. In this example, the Kalman filter operates in
two distinct phases; a) Predict phase, b) Update phase

3.1. Predict Phase

Table 2. Use of the state estimation for Figure 1 in the case
The last position of the vehicle changes in accordance with
the laws of motion in physics and accelerator pedal as well as where there is only information from 2 buses
steering wheel position, and not only a new location estimation
is calculated but also a new covariance is achieved.

3.2. Update Phase

A measurement of the position of the vehicle is obtained by
GPS that is along with uncertainty. Covariance obtained from
the previous step will determine how much it affects the location
obtained from the update phase. In simple terms, if the estimate
is so that it goes away from the actual location, GPS returns it to
the actual position. The interesting point here is that if the
dynamics of the system is not modeled carefully, the efficiency
of this model considerably reduces. In other words, because this
method uses probabilistic models and since these laws have
come into existence form the phenomena in nature the data
obtained from simulation of dynamic system should be used.
As can be seen in the earlier parts no simulation has been
done, but at the end results are presented using data obtained
from simulation of system dynamics. The filter is a recursive
estimator. It means that it uses only previous step and the
In equation (5) Wk~N (0, Q) is easily obtainable by variance
measurement for estimating the current position. Location and
definition as follows:
vehicle speed in linear state space is defined by following
X k = Fx k −1 +W k (5)
equation:
The Q value is obtained through equation (6) as follows:
ªx º
Xk =«.» (1) ª Δt 4 Δt 3 º
«x » « »
¬ ¼ Q = GG T σ a2 = « 2 2 »σ 2
(6)
It is assumed that the vehicles between k and k-1 under « Δt 3 » a
« Δt »
2
constant acceleration α k with normal distribution and the mean ¬ 2 ¼
value is 0 and standard deviation is σ a . Equation (2) is obtained At each time step, measured values (equation (7)) are along
from the law of motion as follows: with noise, and it is assumed that vk has normal distribution with
X k = Fx k −1 + Gak (2) a mean of 0 and standard deviation of σ x .
where F and G can be achieved of equations 3 and 4: z k = Hx k + v k (7)
ª1 Δt º where in equation (7), H=[1 0]. Then by knowing each state
F =« » (3) we go to the next state. This information can be assuming as
¬0 1 ¼
ª0 º ª0 0º
ª Δt 2 º xˆ 0 0 = « » and P0 0 = « » . It should be noted that for
0
¬ ¼ ¬0 0¼
G = «« 2 »» (4)
«¬ Δt »¼

810
uthorized licensed use limited to: Indian Institute Of Technology (Banaras Hindu University) Varanasi. Downloaded on November 03,2024 at 17:12:07 UTC from IEEE Xplore. Restrictions apply
 simulation the equation of constant acceleration for linear T ª Pk k −1 0 º
x ka k −1 = ª xˆ Tk E ªv Tk º º ; Pka k −1 = « » (14)
1
motion is used as x = at 2 . ¬ k −1 ¬ ¼¼
¬« 0 R k »¼
2
x k0 k −1 = x ka k −1
4. Nonlinear Kalman Filter
x ki k −1 = x ka k −1 + §¨ ( L + λ ) Pka k −1 ·¸ i = 1,....L
© ¹i (15)
Kalman filter is provided primarily for linear modes but it is
§ ·
x ki k −1 − ¨ ( L + λ ) Pk k −1 ¸
used in non-linear modes as follows: At each time step, the a a
k −1 = x k i = L + 1,....2 L
Jacobian matrix is calculated according to the predict phase. © ¹i − L
Now, this matrix can be used in the Kalman filter; by this Here an  point obtained from (15) is given to equation (16)
method linearization can be performed around the estimation as follows:
point. But it should be noted that the calculation of Jacobian
matrix is very time consuming for complex functions and also γ ki = h (x ki k −1 ) i = 1,....2L (16)
when in the equation (8), presented as below, two functions of f
and h are highly nonlinear, linearization leads to very poor Now, by applying weighted values in equation (16), equation
performance of filter: (17) can be obtained that is used for covariance and estimated
x k = f ( x k −1,u k ) + w k ; z k = h ( x k ) + v k
measured value as follows:
(8) 2L 2L

¦W ¦W
T
ª i ºª i ˆ º
s γk ; «¬x k k −1 − x k k −1 »¼ ¬γ k − z k ¼
i i i
Unscented Kalman filter (UKF) uses specific sampling zˆk = Px k z k = c ˆ (17)
technique to make a matrix, where  points are chosen; now i =0 i =0
with the use of the information, estimation and covariance of
estimation can be obtained. The interesting point here is that
k k = Px k z k P −1x k z k ; xˆ k k = xˆ k k −1 + K k (z k − zˆk )
(18)
there is no need to calculate the Jacobian matrix. For this filter
Pk k = Pk k −1 − K k Px k z k K Tk
UKF is used.
It can be seen that the initial values for next state are
4.1. Predict Phase obtained according to outputs of equation (18).
Two equations are converted to equation (9) to estimate
position and covariance as follows: 5. Simulation Results
T ª Pk −1 k −1 0 º
x ka −1 k −1 = ªxˆTk −1 k −1 E ªw Tk º º ; Pka−1 k −1 = « » (9) In this section the simulation results of this study are
¬« ¬ ¼ »¼ «¬ 0 Q k »¼
presented using linear Kalman filter and non-linear Kalman
=2L+1 points is calculated by equation (10) and L is added filter at sample power system.
to the matrix, as follows:
x k0 −1 k −1 = x ka −1 k −1 1000 800

800 Created of Value Measure Without Noise.

x ki −1 k −1 = x ka −1 k −1 + §¨ (L + λ )Pka−1 k −1 ·¸ i = 1,....L
Value of M easurement

600
(10) 600
© ¹i 400
400

x ki −1 k −1 = x ka −1 k −1 − §¨ (L + λ )Pka−1 k −1 ·¸ i = L + 1,....2L 200

© ¹i − L
200

0 0

In equation (10), §¨ (L + λ )Pka−1 k −1 ·¸ the aim is to select ith

0 50 100 150 200 250 300 350 400 0 50 100 150 200 250 300 350 400
Number of Measure Number of Measurement

© ¹i (a) (b)
Created of Value Measure by Considering Nois

1000 1000
column of matrix. Using  points obtained and using it non-
Augmentedof A,B,C parts.

linear equation f, equation 11 can be obtained as follows: 500 500

x ki k −1 = f (x ki −1 k −1 ) i = 0...2L (11)
0 0
The weighted  points are combined for prediction phase and Valuof MeasureWith Noise.
Valueof MeasureWithout Noise.
equation (12) is obtained as below: -500 -500
Prediction of ValueMeasurement By Filter Kalman.
0 50 100 150 200 250 300 350 400 0 50 100 150 200 250 300 350 400
2L Number of Measurement Number of Measurement

xˆ k k −1 = ¦W
i =0
s
i
x i
k k −1
(12)
(c) (d)
Fig. 3. Using a linear Kalman filter, (a) estimated values of
2L measurer using linear Kalman filter, (b) measured value
¦W
T
Pk = i ªx i − xˆ k º ªx i − xˆ k º produced without taking into account the noise on the measured
k −1 c «¬ k k −1 k −1 »¼«¬ k k −1 k −1 »¼
i =0 value,(c) measured value produced by taking into account the
Typical values for the parameters of (13) are as follows: noise on the measured value, and (d) shows (a), (b) and (c) at
the same time
λ λ
W si = ; W ci = + (1 − α 2 + β )
L +λ L +λ (13) 5.1. Linear Kalman Filter
1
W si = W ci = ; λ = α 2 (L + κ ) − L Figure 3 shows the performance of linear Kalman filter; (a),
2( L + λ ) estimated values of measurer using linear Kalman filter, (b)
measured value produced without taking into account the noise
4.2. Update Phase on the measured value(c)measured value produced by taking
The difference of this section with the previous section is that into account the noise on the measured value and (d) shows a, b
here the mean and covariance of the measurement noise is used, and c at the same time. Filter acts excellent for small values but
which is shown in equations (14) and (15): for larger ones it does not operate properly; it seems as

811
uthorized licensed use limited to: Indian Institute Of Technology (Banaras Hindu University) Varanasi. Downloaded on November 03,2024 at 17:12:07 UTC from IEEE Xplore. Restrictions apply
 mentioned earlier, this model does not provide correct answers results are shown for both linear and nonlinear systems. The
in cases where the system is not dynamically modeled. estimation method used for linear and non-linear systems are
respectively linear and nonlinear Kalman filters.
5.2. Non-linear Kalman Filter
Figure 4 shows non-linear Kalman filter which respectively 7. References
include, (a) estimated values of measurer using non-linear
Kalman filter, (b) value produced by measurer without taking [1] F. M. Shahir and E. Babaei, “Evaluating the Dynamic
into account the noise on the measurer, (c) value produced by Stability of Power System Using UPFC based on Indirect
measurer with taking into account the noise on the measurer, (d) Matrix Converter,” JOACE, Vol. 1, No. 4, pp. 279-284,
shows a, b and c at the same time. As seen in Figure 4, it seems Dec., 2013.
that this method was able to eliminate the noise at an acceptable [2] F. M. Shahir and E. Babaei, “Evaluation of Power System
level. IEEE 9-bus 3 generators system implemented in Stability by UPFC via Two Shunt Voltage-Source
DigSilent. The effect of non-linear Kalman filter is examined Converters and a Series Capacitor,” in Proc. ICEE, 2012,
pp. 318-323.
fairly. Synchronous 2nd generator curves obtained is shown in [3] A. Kumar Singh and B.C. Pal, “Decentralized dynamic
Fig. 5. state estimation in power systems using unscented
transformation,” IEEE Trans. on Power Systems, vol. 29,
Creadated of Value Measure Without considering

1
1
no. 2, pp. 794-804, 2014.
With Unscented Filter Kalman.

[4] E. Babaei, A. Zackery, A. M. Motlagh , and, F. M. Shahir,

Predicted of Value M easure

0.5
0.5
“Design of Robust Power System Stabilizer based-on
0 0
Optimal Control Theory,” in Proc. ICTPE, 2013.
-0.5 -0.5
[5] F. M. Shahir and E. Babaei, “Assessment of Power System
Stability by UPFC with Two Shunt Voltage-Source
-1
0 100 200 300 400 500 600 700 800 900 1000 -1
0 100 200 300 400 500 600 700 800 900 1000 Converters and a Series Capacitor,” ERR, Vol. 1, No. 4, pp.
Number of Measurement Number of Measurement
104-115, October 2013
(a) (b) [6] F. M. Shahir, M. Farsadi, H. Zafari and A. Sadighmanesh,
Created of Value Measure with considering noi

2 2 Prectidedof valueMeasureByUnscentedFilter Kalman

Valueof MeasureWithout Noise.
Valueof MeasureWithNoise.
“Solving Economic Emission Load Dispatch Problems
augmented of A,B,C Parts.

1 1 Using Particle Swarm Optimization with Smart Inertia

Factor,” in Proc. ELECO, 2015, pp. 500-504.
0 0
[7] Y. Xu, “A reliable intelligent system for real-time dynamic
-1 -1
security assessment of power systems,” IEEE Trans. on
Power Systems, vol. 27, no.3, pp. 1253-1263, 2012.
-2
0 100 200 300 400 500 600 700 800 900 1000
-2
0 100 200 300 400 500 600 700 800 900 1000 [8] D.Z. Fang, “Improved hybrid approach to transient stability
Number of Measurement Number of Measurement
assessment,” IEE Proceedings-Generation, Transmission
(c) (d) and Distribution, vol. 152, no. 2, pp. 201-207, 2005.
Fig. 4. Application of non-linear Kalman filter, (a) estimated [9] N. Fernandopulle and R.TH. Alden, “Improved dynamic
values of measurer using non-linear Kalman filter, (b) value security assessment for AC/DC power systems using energy
produced by measurer without taking into account the noise on functions,” IEEE Trans. on Power Systems, vol. 18, no. 4,
the measurer, (c) value produced by measurer with taking into pp. 1470-1477, 2003.
account the noise on the measurer, (d) shows (a), (b) and (c) at [10] M.A. El-Kady, “Dynamic security assessment utilizing the
the same time transient energy function method,” IEEE Trans. on Power
Systems, vol. 1, no .3, pp. 284-291, 1986.
[11] E. Ghahremani and I. Kamwa, “Dynamic state estimation
in power system by applying the extended Kalman filter
with unknown inputs to phasor measurements,” IEEE
Trans. on Power Systems, vol. 26, no. 4, pp. 2556-2566,
2011.
[12] W. Miller and J. Lewis, “Dynamic state estimation in
power systems,” IEEE Trans. on automatic control, vol.
16, no. 6, pp. 841-846, 1971.
[13] G. Valverde and V. Terzija, “Unscented Kalman filter for
power system dynamic state estimation,” IET Generation,
Transmission & Distribution, vol. 5, no. 1, pp. 29-37,
2011.
[14] S. Wang, W. Gao and A.P.S. Meliopoulos, “An alternative
method for power system dynamic state estimation based
on unscented transform,” IEEE Trans. on Power Systems,
Fig. 5. Synchronous 2nd generator curves obtained of IEEE 9- vol. 27, no .2, pp. 942-950, 2012.
bus and 3 gen. [15] D. Nazarpour and F. M. Shahir, “Analysing THD and
Coordinate Control for Power System and DG based on PI-
Controller," in Proc. ICNRAECE, 2016.
6. Conclusion [16] M. Farsadi, F. Nazari Heris, F. M. Shahir and A.
Sadighmanesh, “The Role of The Intelligent
Given the importance of estimation in the power systems to Reconfiguration of distribution Network on Reduction The
obtain important information such as pharos voltage and current Energy not Supplied Costs in The Electricity Market
of branches in static state or dynamic state estimation of power through Case Studies Using Software NEPLAN and
systems, such as angle of rotor of synchronous machine and so DIGSILENT Power Factory,” in Proc. ELECO, 2015, pp.
on, the need for a proper estimation method is felt. The method 445-447.
should be adapted to the conditions of system and characteristics [17] J .Wood and B.F. Wollenberg, Power System Generation,
of the estimators. The feature of the system is its non-linearity Operation and Control, Wily Interscince publication, 1996.
and feature of measurements is the probability of its values.
Kalman Filter can adapt to these features and has accurate
answers resistant to noise at the right time. The simulation

812
uthorized licensed use limited to: Indian Institute Of Technology (Banaras Hindu University) Varanasi. Downloaded on November 03,2024 at 17:12:07 UTC from IEEE Xplore. Restrictions apply