2015 IEEE Power, Communication and Information Technology Conference (PCITC)

Siksha '0' Anusandhan University, Bhubaneswar, India.

A Robust Extended Kalman Filter for the

Estimation of Time Varying Power System
Harmonics in Noise
Pravati Nayak B.N. Sahu
Departmetn of Electrical Engineering Department of Electronics & Comm. Engg.
Siksha 0 Anusandhan University, Siksha 0 Anusandhan University
Bhubaneswar, India Bhubaneswar, India
pravatinayak@soauniversity.ac.in badrinarayansahu@soauniversity.ac.in

Abstract- Measurement of time varying harmonics has adalines and neural network [5,6], recursive Newton and
assumed great importance for power quality estimation in Gauss-Newton type algorithms [7-9]], Kalman filters [10-
micro and smart grid environments. It is well known that the 12], mathematical transformation [13,14], phasor
power signals undergo dynamic transitions during power measurements [15], etc. Apart from these techniques some
quality and fault type disturbances. Also accurate estimation of the time-frequency techniques proposed in recent years
of amplitude, phase and frequency of a sinusoid in the
for power quality and harmonic estimation use the Wavelet
presence of harmonics/inter harmonics and noise plays an
transform [16-18], S-Transform [19], besides the Gabor
important role in wide variety of power system applications,
transform, and the Wigner-Ville function (WDF).
like protection, power quality control and state monitoring.
One of the widely used approach like the Fourier linear Extended Kalman filter (EKF) is basically a powerful
combiners become ineffective to track the sudden changes in estimator keeping in view of state estimations. Optimum
the power harmonic amplitude and phase angles in the value of mean and covariance in terms of minimum mean
presence noise and distortions. This paper, therefore, square error can be obtained by the EKF for linear systems.
presents a new adaptive robust Kalman filter (REKF) for In the case of EKF, the probability distribution fimction
estimating the fundamental and harmonic phasors and
(pdf) is propagated linearly around the operating point at
frequency in a distributed generation systems suitable for
each instant of time and the approximation error between
the detection of islanding, non islanding and switching events.
the Iinearised model and the original nonlinear system may
The REKF is based on linear H-infinity technique to avoid
become significant when there are large deviations of the
the error divergence due to linearization and Jacobian
estimated state from the real nonlinear one, particularly
formulation in the usual extended Kalman filter (EKF)
.Further the noise covariance matrices of the robust Kalman
when the signal contain time varying harmonics. Thus an
filter are tuned iteratively to produce faster convergence and
adaptive robust extended Kalman filter based on certain H­
speed of accuracy in tracking time varying harmonic infinity filter concepts is presented in this paper to provide
phasors. a more accurate estimation of time varying harmonic
components obtained from power system network. The
Keywords-harmonics; Kalman; linearization; Jacobian; principles and algorithms of the EKF are described as
covariance. follows:

I. INTRODUCTION II. POWER SYSTEM SIGNAL MODEL

The abundant utilization of power electronics based The observed power, voltage or current signal is assumed
devices in power systems gives rise to harmonic pollution to comprise a decaying dc component, fundamental and
resulting in severe power quality problems. Further, many harmonics of order N and is represented by,
adverse effects like more losses in power distribution y(t) set) + v(t)
networks, overheating of electric drives, malfimction of
=
(1)
relays and circuit breakers , severe waveform distortions, where yet) is the time-varying voltage or current signal at
etc. are the consequence of harmonic pollution. Thus the
time t, vet) is the random noise and s(t) is the signal
voltage and current waveforms in the above cases may
comprise a combination of fundamental frequency, integral model given by,
N

2 )a"cosnWof + b"sinnwot)
harmonics, and high frequency transients. Besides due to -aocl
power network disturbances nonstationary time varying s(t)= adee + (2)
11=1
integral harmonics with varying amplitude and phase
angles can occur, requiring their dynamic estimation. Thus where an and b n are the direct and quadrature
it is essential that the harmonics generated in a power
network should be monitored by estimating its amplitude components of the nth order harmonic, ade and a de are
and phase in a high noise environment. Various estimation the magnitude and time constant of the decaying dc
methods have been proposed: Fast Fourier transforms [1], component.
least square error technique [2,3], adaptive notch filter [4],

978-1-4799-7455-9/15/$31.00 ©2015 IEEE

 2015 IEEE Power, Communication and Information Technology Conference (PCITC)
Siksha '0' Anusandhan University, Bhubaneswar, India.

The discrete version of the above model is obtained after

expanding the decaying dc term as Ck1k-1 =
Pk1k.1, for Cok > � C O,k, � > 0, (12)
N
2 l
s(k)= adc - adcak + I (an cosncdc + bn sinnwk) (3) and Ck1k-1 =
(PkiLI - L�Lk I y r otherwise
11=1
(13)
where a = a deTs, OJ = OJoTs, and Ts is the sampling

interval and is equal to 1 I is ; is is the sampling The value of C O k is adapted as follows:

frequency. The principle and algorithm of the EKF for -
For k=0, C O,k e; ,
estimating the parameters of the signal shown in eq.(3) are =

described as follows: -
pC O,k -I +e;
and for k>O C Ok (14)
A. Ro bust Extended KalmanFilter (REKF) p +l
The value of p is chosen as 0.98 and is a forgetting factor.
The complete discrete version of the robust extended
Kalman filter is described below: Thus the filter stability is guaranteed and the error
A A
divergence is bounded by using this formulation.
X klk-I =f( X k-I )+ Wk (4) However, statistical properties of the process and
measurement noise covariances Q, and R are needed to
zk=h(Xk )+Vk (5)
be known for the successful implementation of the robust
Xk = [adc ,a ,al , b l' a2, b 2,······· ..········,an, bn] T
a extended Kalman filter. The convergence of the robust
EKF can be degraded if these parameters are either known
zk = [1,k" com,sinkw,cos2kw,sin2kw " ....... ""'
apriori or are adapted depending on the error at every
coskrw,sinlaw] xXk iteration.
T The innovation error sequence is written as
�lk.I =Fk Pk.IFk +Q k (6) A

(15)
and xk1k_1 = E[wk1k_IJ ,
Taking the variance on both the sides of the above
A A
T -I equation, the expression for error covariance is obtained as
Pk1k-1 = E[(Xk - Xklk-I )(Xk - Xk1k_l) S 1, (7) T
+ Rk
=

&>0 ee,k =
HkPklk-1Hk (16)

Qk = ql R r , 1 is an unit matrix of appropriate The error covariance is rewritten as

=
k
2
L
A

dimension ee,k = E(e;), and e e,k = > i Ik

and the error covariance matrix C is given by i=1
2 l
Ck1k_1 =
(PkiLI - L�Lk I y r (8) The measurement noise covariance Rk is thus obtained
from equation (26) as
Of(x)
where Fk - I ,and L k is an unity matrix
ax W=Wk_J (17 )
of appropriate dimension. Here y is a tuning parameter
In a similar way the process noise error covariance is
>0, and Ck1k-1 >0. obtained as
k
The Kalman filter gain is obtained as
A

K k = Ck1k_1H/ (HkCklk_1H/ +RkT

I
(9)
Q =
QJXk -X�k-l)(\:k -xkjk_l+� -F�k-lF;}/k
i=1

1\ 1\ 1\
(18)
Xk Xklk-l + KJZk -14 Xklk-l )
= (10)
Further the equation ( 18 ) is approximated as

-1 T -1 -1 (19 )
D
r k =
( Cklk-l + Hk Rk Hk ) (11)

Fk 122x22=
III. S[MULAnON REsULTS

Hk [l,k"coskOJ,sin kOJ,cos2kOJ,sin 2kOJ"...........,

= The following case studies are performed for
computing the fundamental and harmonic components with
cos knOJ,sin knOJ ]
time varying amplitude and phase angles. A sampling rate
Further to improve the stability of the filter and keep the of 5.0 kHz is chosen to include a large number of
estimation error within a reasonable boundedness the harmonics if required.
following adaptive variation is used:
 2015 IEEE Power, Communication and Information Technology Conference (PCITC)
Siksha '0' Anusandhan University, Bhubaneswar, India.

A. Estimation of Time varying Fundamental and

Harmonics
The test signal comprising harmonic components up to 19th
are represented as given below (amplitudes are in per unit
and phase angles are in radians):
V(t} 1.2 x cos{wt + O.S) + O.4x cos{3wt + 0.7S) +
=

0.3x cos{Swt + O.S) + 0.1x cos{7wt+ 0.6) (20)

+ 0.08x cos{l lwt+ 0.2) + 0..02x cos{l3wt + 0.1)
+ O.OOSx cos{19wt + 0.7) + ({t ) (e)

5
!o--
-2_ O --c'�00�2
�00��30
C
CO ----040�
---c5
- 0 0
� -o
0-�
6 OO�-c7
C00-� o�8 OO�90�0�' cc!0.OO (f)
Samples

'" (a)
'C
�

:a.

� :: 8�_____ ----,1 r' \:\,l,-,.�,-------1

I
.2 1
'C
'"

� O. 8 L-"'-___'_:_:_'_:_---c'--c--...L_:_'_:_---c'--c-�- c'----c"-:--..c'
� 0 100 200 300 400 500 600 700 800 900 1000
w

�
'"
0 .8 rr----.-....-----,
-- ---,-....-----,--,--,---
-- ----,---,
(g)

I �i\!+'_--------c'�!\,
Q.

06

f",\ �, ------1
iE i:
-.

4
i I
.TI
� 02�
.
0
�
' ·___'_:_:_'_:_---c'_-...L_:_'_:_---c'_-L_c'----c"_..c'
100 200 300 400 500 600 700 800 900 1000

(b)
Samples

(h)

/:
I

(c) Frequency
2
. !o--
0 --c,�00 =-0----,c!
""0...., 0,...3�00�""40
,...
2 ..., CO-----='
O
50 60
==
-O ----C0
70O----O
"" 8=00
--
O --,� �
9 001000
0.55 !
(i)
Fig. I Estimation of Fundamental and Harmonic Components

The frequency of the fundamental component is varied

from SO Hz to 49 Hz for the sample count SOO<k<700.
During this period the fundamental component is increased
to 1.44 pu, the third harmonic component to 0.48 pu, the
fifth harmonic component to 0.36 pu. Also the
(d)
 2015 IEEE Power, Communication and Information Technology Conference (PCITC)
Siksha '0' Anusandhan University, Bhubaneswar, India.

corresponding phase angles of the fundamental, third

harmonic, fifth harmonic are changed to 0.6,0.85, and 0.3,
respectively. Fig.1 shows the actual time varying
harmonically distorted signal mixed with white Gaussian
noise of SNR=30 dB and estimated fundamental amplitude
and phase along with some selected harmonic components.
From the results shown in Fig.1 it is observed that the
tracking of the true values is obtained within one
fundamental cycle of the signal waveform.

The estimation errors are indicated using two

statistical operations on time domain error vectors -
Mean Absolute Error (MAE) and Root Mean Square Error
(RMSE). They can be shown as given below.

MAE f IA;-E ;1 (21)

;�l N
N

RMSE = N -2 L (A;-E;Y (22)

i=! ·'::-0 ----c=------O;;;;-
; ---;3;;
OO;-
;; -
- -;;;-
;;;- ----�
-;; -�

where N is the length of the signal estimated, A; is the Cc"r

p,es

actual magnitude of the /" sample of the signal and E; is the

--Onginal
----�--- Estimated

estimated magnitude of the same sample. Table-1 exhibits

2_5 .

both MAE and RMSE from which it can be observed that

the error amplitudes are found to be very small for the
fundamental and harmonic components. However, the
phase angle errors are found to be higher for the harmonic
components as can be seen from Table-I.
TABLE I. ERROR IN ESTIMAnON USING REKF
HARMONIC ANALYZER FOR CASE 2 Samples

(d)
Property Harmonic MAE RMSE
2_5 '

1st 0.0037 0.0072

3rd 0.0042 0.0058
5th 0.0032 0.0049
Amplitude
7th 0.0042 0.0058
lIth 0.0078 0.0089
13th 0.0222 0.0225
1st 0.0040 0.0222 Samples

3rd 0.0176 0.0337 (e)

Phase
5th 0.0200 0.0274 1------- �:��::ed

7th 0.0167 0.0226

lIth 0.0220 0.0266
13th 0.0277 0.0297

Samples

(f)
Fig. 2. Estimation of fundamental and hannonic components of damped
sinusoids.

B. Estimation of Damped Sinusoids with Harmonics

The damped sinusoidal signal with harmonic distortion is
Samples represented in discrete form as
(a)
 2015 IEEE Power, Communication and Information Technology Conference (PCITC)
Siksha '0' Anusandhan University, Bhubaneswar, India.

TABLE [[ ERROR VALUES OF HARMONIC ANALYZERS FOR CASE 2

Property Harmonic LMS RDFT EKF REKF

1 st O. 0062 0. 0035 0.0046

0.0379
O.
2nd O. 0068 0. 0030
0.0306 0048
Amplitude
O. O.
3rd O. 0069 0. 0037
0062 0052
O. O.
5th O. 0083 0. 0047
0055 0062
O.
1 st 0. 0061 0. 0015
0.0167 0044

2nd 0.0173 0. 7872 0.0125

I -- -- �:��::ed
-

0. 1406
Phase
3rd 0.0229 0. 3955 0.0174
0. 1352

5th 0.0410 0. 1311 0.0300

0. 1081

Yen) = al(n) sin(w(n)n + phl(n» +

a2(n)sin(2 w(n)n + ph2(n» + a3(n)sin(3 w(n)n +

ph3(n» + a 5(n)sin(5 w(n)n + ph5(n»
(23) I --- --- �:��:!ed
-

where w(n)=2*pi*f(n)*dt; a1(n)=A1*exp(-alfa1 *n*dt),

a3(n)=A3*exp( -alfa1 *n*dt),a5(n)=AS*exp( -alfa1 *n*dt),
a2(n)=A2*exp( -alfa1 *n*dt),adc(n)=AO*exp( -n*alfadc*dt),
ph1(n)=pi/8,ph3(n)=pi/6,phS(n)=pi/4,ph2(n)=SO*pi/ 180,A
o = O.S, Al = 1.S, A2 = .8, A3 = 0.2S,AS =O. lS The
damping factor the dc component alfadc= lO, and for the
sinusoids belonging to the fundamental and harmonics is Samples
(c)
chosen as alfa1 =6.0. These damping factors are quite
reasonable for practical signals. The damped sinusoidal I --- --- �:��:!ed
-

signal is sampled with a sampling frequency of 2 kHz and

is mixed white Gaussian noise having SNR=30dB. The
proposed REKF algorithm is applied to estimate the
parameters of the damped sinusoidal signal with decaying
dc and harmonics and Fig.2 exhibits the tracking of the
various components. From the results it is quite obvious
that the filter converges rapidly within 2 cycles. Further
this is corroborated from Table 2 where both the amplitude
and phase errors for the fundamental and harmonic
components are given. The fundamental amplitude and
1--- ____ �:��:!ed
phase angle errors are found to be 0.061% and 0.062%,
respectively. 3;

The magnitude of the fundamental amplitude is then ,i

changed from 1.Spu to 1.2 pu between 100 and 3S0 J
samples, whereas the third harmonic amplitude is
increased by 20% between 200 and 2S0 samples. Fig.3
shows the original time varying damped sinusoidal signal -1 0!"-
-�-----O�--3�OO�-�-�cc---o!
with harmonics and decaying dc component along with the (ejP'-
tracked amplitudes of fundamental, 3rd harmonic, and dc
Fig. 3. Estimation of time varying damped sinusoids.
component. Also Fig.3(c) exhibits the fundamental phase.
From the figures it is observed that even with significant
signal parameter changes the proposed REKF tracks the
parameters of the time varying signal more or less
accurately.
 2015 IEEE Power, Communication and Information Technology Conference (PCITC)
Siksha '0' Anusandhan University, Bhubaneswar, India.

IV. CONCLUSION [8] Vladimir V. Terzija, Vladimir Stanojevic, "Two stage Improved
Recursive Newton-Type Algorithm for Power-Quality Indices
The paper presents a robust extended Kalman filter for Estimation" IEEE Transactions on Power Delivery, Vo1.22, No.3,
tracking time varying fundamental, harmonics, and 2007, pp. 1351-1359.
superimposed decaying de components accurately within a [9] A.M. Alkandari, S.A. Soliman, "Measurement of a power system
time frame of 2 cycles based on a 50 Hz power system nominal voltage, frequency, and voltage flicker parameters",
network. The REKF utilizes the robustness property of the International Journal of Electrical Power& Energy Systems, Vol.3l,
No.7-8, 2009, pp. 295-30 I.
H-infinity filter that minimizes the error divergence due to
linearization and Jacobian formulation in the conventional [10] P.K. Ray, B.Subudhi, Ensemble Kalman filter based power system
harmonic estimation, IEEE Trans. on Instrumentation and
extended Kalman filter algorithm. Also the paper presents Measurement, Vo1.61, No. 12, 2012, pp. 3216-3224.
adaptive algorithms for updating the process and
[11] J. Reddy, P. K. Dash, R. Samantaray, and A. K. Moharana, "Fast
measurement noise covariance matrices in a recursive tracking of power quality disturbance signals using an optimized
manner. Several case studies reveal that the proposed unscented filter, IEEE Trans. Instrum. Meas., Vol. 58, No. 12, 2009,
robust filter can successfully estimate the time varying pp. 3943-3952.
power system harmonics accurately. Comparison with [12] C.I. Chen, G. W.Chang, R. C. Hong , H. M. Lee," Extended real model
other convention harmonic tracking algorithms provides of Kalman filter for time- varying harmonics estimation", IEEE
the superiority of performance of the proposed filter in Transactions on Power Delivery, Vol. 25, No. 1, 20 I 0, pp. 17-26.
terms of convergence speed and accuracy. [13] F.A. S. Neves, M. C. Cavalcanti, G.M. S. Azevedo, J.P. de Aruda, "A
method for extracting the fundamental frequency, positive sequence
voltage vector based on simple mathematical transformations", IEEE
REFERENCES Trans. on Industrial Electronics, Vo1.56, No. 5, 2009, pp. 1539-1547.
[1] T. X. Zhu, Exact harmonicslinterharmonics calculation using [14] M.A. P. Garza, J.A Serna, "Dynamic Phasor and Frequency
adaptive window width, IEEE Transactions on Power Delivery, Estimates through Maximally Flat Differentiators", IEEE Trans.
Vo1. 22, No.4, 2007, pp. 2279. on Instrumentation and Measurement, Vo1.59, No.7, 2010, pp.1803-
1811.
[2] M.Bettayeb, and U. Quidiani, A hybrid least squares-GA based
algorithm for harmonic estimation, IEEE Trans. on Power Delivery, [15] R.A.Zadeh, A.Ghosh, G.Ledwitch, F.Zare, "Analysis of phasor
Vo1.18, No.2, 2003, pp. 52-67. measurement method in tracking power system frequency of
distorted signals", lET Gen., Trans. , Distri. , Vol..4, No.7, 2010,
[3] S. Mishra, Hybrid least squares adaptive bacterial foraging strategy
pp.759-764.
for harmonic estimation , lEE Proc:-Gener. Transm. Distrib. ,
Vol. 152, NO. 3 ,2005, pp. 379-389. [16] J.Barros, R. I. Diego, M. de Apraiz, Applications of wavelet
transform for analysis of harmonic distortions in power systems: A
[4] M. Mojiri, M.K. Ghartemani, A. Bakhshai, "Processing of
review, IEEE Trans. on Instrumentation and Measurement, Vo1.61,
Harmonics and Interharmonics Using an adaptive Notch Filter",
No.l0, 2012, pp.2604-2611.
IEEE Transactions on Power Delivery, Vo1.25, No.2, 2010, pp. 534-
542. [17] N.C. F. Tse, Y.C. Chan, W-H Lau, L. L. Lai, Hybrid Waveelet and
Hilbert Transform with frequency shifting decomposition for power
[5] H. C. Lin, Intelligent Neural Network-based fast power system
quality analysis, IEEE Trans. on Instrumentation and Measurement,
harmonic detection, IEEE Transactions on Industrial Electronics,
Vol. 61, No. 12, 2012, pp.3225-3233.
Vol. 54, No. 1, 2007, pp. 43-53.
[18] J. Ren, M. Kezunovic, Real-Time Power System Frequency and
[6] G. W. chang, Cheng-I Chen Q.W. Liang, "A two-stage adaline for
Phasor Estimation Using Recursive Wavelet Transform, IEEE Trans.
harmonics and interharmonics measurement". IEEE Trans. on
on Power Delivery, Vo1.26, No.3, 2011, pp. 1392-1402.
Industrial Electronics, Vo1.56, No.6, 2009, pp. 2220-2228.
[19] P.K.Dash, M. V.Chilukuri, Hybrid S-transform and Kalman filtering
[7] Jung Zheng, Kenneth W.K.Lui, W.K.Ma, and H. C. So "Two
approach for detection and measurement short duration
Simplified Recursive Gauss-Newton Algorithms for Direct
disturbances in power networks, IEEE Trans. on Instrumentation
Amplitude and phase tracking of a real Sinusoid". IEEE Signal and Measurement, Vo1.53, No.2, 2004, pp. 588-596.
Processing letters, Vol. 14, No.12, 2007, pp. 972-975.