Download as pdf or txt
Download as pdf or txt
You are on page 1of 6

Harmonic Estimation in a Power System using

Hybrid H∞-Adaline Algorithm


1
H.K.SAHOO, 2P.K.DASH, 3N.P.RATH, 4B.N.SAHU

14 2
Asst.professor,Silicon Institute of Technology, Director PG& Research, Silicon Institute of Technology,
Bhubaneswar, Orissa, India Bhubaneswar, Orissa, India
3
Reader&Head, UCE, Burla, Orissa, India

Abstract- Accurate computation of harmonics is really a when estimating multiple frequency components(n) because of
challenging problem in power system. Many algorithms have large number of parameters(2n) that had to be identified .Since
been proposed for harmonic estimation. In this paper a novel amplitudes and phases are different quantitatively, it is
hybrid approach is proposed to estimate the amplitudes and difficult to get the homogeneous genetic pool with respect to
phases of fundamental, third and fifth harmonics of three
the final solution. In this paper, a nonlinear filter known as H∞
different types of distorted power signals in presence of white
noise. A robust estimator known as H∞ Filter is used for filter has been proposed for estimating the amplitudes [2] and
amplitude estimation, which is based upon state space modeling Neural network based Adaline structure is used to estimate
of the signals and an Adaptive Linear Combiner called ‘Adaline’ phases of different harmonics [9] of a distorted power signal in
based on the Neural network, which is very simple and faster is presence of white noise. H∞ filter is applied to a state-space
used for phase estimation. The simulation results are compared representation of noisy power signal modeled under the
with that of Extended Kalman Filter to show the tracking assumption that the number of sinusoids is known.
capability of the proposed algorithm.
II. H∞ FILTERING APPROACH
Index Terms: - Adaline, Extended Kalman Filter, H∞ Filter,
Neural Network, White Gaussian noise
H∞ filter is an optimal dynamic estimator and is suitable for
I. INTRODUCTION describing state variables in a power system. Its estimated
process is divided into state prediction and state filter. The
T HE wide spread applications of electronically controlled
recursion process of the filter through linearization for
estimating the parameters of power signal can be obtained as
loads have increased the harmonic distortion in power system follows:
voltage and current waveforms. Power quality has been an
issue of growing concern amongst a broad spectrum of power State Prediction:
customers over the past few years. The increasing use of solid
state devices such as diodes and thyristor rectifiers, lighting xk +1|k = f ( xˆk |k ) (1)
equipments, uninterruptible power supplies, arc furnaces,
adjustable speed motor drives etc has brought about an P k + 1| k = Fk P k | k − 1 Fk* + G G*kT
increase in the distortion of voltages and current waveforms in k
industrial, commercial and residential power networks. ⎡H ⎤
k⎥P
(2)
− Fk P k | k − 1 [H *T L*kT ]R e−,1k ⎢ k | k − 1 F *kT + Qk
Moreover, starting of large loads either on the affected site or k ⎢⎣ L k ⎥⎦
by a consumer on the same circuit and faults on other branches
of the network can cause voltage dips.
In order to provide the quality of the delivered power, it
is imperative to know the harmonics parameters such as ⎡H ⎤
k⎥P *T
magnitude and phase. This is essential for designing filter for R e , k = R0 + ⎢ k | k − 1 [ H k L *kT ] (3)
L
⎢⎣ k ⎥⎦
eliminating or reducing the effects of harmonics in a power
system. Many algorithms are proposed to evaluate the
harmonics [6,8] in a power system. Recently developments of ⎡I 0 ⎤
Where (4)
soft computing techniques have encouraged the researchers to R0 = ⎢ ⎥
⎢0 −γ2 I⎥
use these methods for harmonic estimation. Since estimation ⎣ f ⎦
of harmonic parameters is a nonlinear problem, Genetic
Algorithm (GA), as Qidwai and Betayeb used heuristic and In (1), the symbols ~ and ^ stand for the estimated and
stochastic global searching algorithm for this purpose [6]. But predicted values, respectively.
GA suffers from larger time requirement for convergence
978-1-4244-4547-9/09/$26.00 ©2009 IEEE 1 TENCON 2009
In case of H∞ filter applied to a linear system, the modified Where
equations will be
a1 = 0.15sin 2π f1 kTs + 0.05sin 2π f 5 kTs
P k + 1| k = F Pk | k − 1 F * + β 2 G G T −
k k a3 = 0.05sin 2π f 3 kTs + 0.02sin 2π f 5 kTs
⎡H ⎤ (5)
F Pk | k − 1 [ H * T H * T ] R e− ,1k ⎢
k ⎥
Pk | k − 1 F *k T + Q k a5 = 0.025sin 2π f1 kTs + 0.005sin 2π f5 kTs
k ⎢⎣ L k ⎥⎦

Where f1 = 1Hz , f3 = 3Hz , f5 = 6 Hz , α = 5

A) Amplitude Estimation Using H∞ Filter:


Pk−|k1 = Pk−|k1−1 + (1 − γ −f 2 ) H *T H > 0 ( L = H ) (6)
k k Consider a signal yk at time k is a sinusoid zk in the presence
of white Gaussian noise vk.
State Filter:
yk = zk + vk (11)
xˆk |k = xk |k −1 + K k ( yk − Hxk |k −1 ) (7)
Where
The filter gain Kk is calculated to be
zk = a1 sin ( kω1Ts + φ1 ) (12)

K = P k | k − 1 H *T ( Rk + H P k | k − 1 H *T ) −1 (8) Where
k k k k

Where ω1 = Fundamental of angular frequency;

xˆk |k = State variable after filtering φ1 = Fundamental of phase angle;

xk |k −1 = State variable after estimation a1 = Fundamental amplitude of the signal.

K k = Gain The observation noise vk is a Gaussian white noise with zero


mean and variance σ v2 . The covariance of measured errors
yk − Hxk |k −1 = Innovation vector. is Rk = E ⎡⎣ vk vk ⎤⎦ , where * means the complex conjugate
∗T

and T is the transpose.


III. SIGNAL MODEL FOR HARMONICS ESTIMATION
For the static signal the state variables are given by:
In this case we have considered a static signal (Test Signal1)
and a dynamic signal (Test Signal 2) for estimation, which
contains higher harmonics of the 3rd, 5th order and a slowly
xk (1) = a1
decaying dc component. xk (2) = a2
The test signal 1 is given by: xk (3) = a3 (13)

zk = 1.2sin ( kwTs + π / 6) + 0.8sin ( 3kwTs + π / 3) xk (4) = aDC


(9)
+0.5sin ( 5kwTs + π / 4) + 0.5exp ( −α kTs ) xk (5) = aDCα DC

The test signal 2 is given by:

z k = (1 .5 + a 1 ( t ) ) s in ( k w T s + π / 6 )
+ ( 0 .5 + a 3 ( t ) ) s in ( 3 k w T s + π / 3 ) (10)
+ ( 0 .2 + a 5 ( t ) ) s in ( 5 k w T s + π / 4 )
+ 0 .5 e x p ( − α k T s )

2
For dynamic signal state variables are given by: Weights are updated using a modified Widrow-Hopf delta rule
as:
xk (1) = 1.5 + a1 ( t )
W ( k + 1) = W ( k ) + α e ( k ) X ( k ) / x kT x k (22)
xk (2) = 0.5 + a3 ( t )
The error signal is given by:
xk (3) = 0.2 + a5 ( t ) (14)
Λ
xk (4) = aDC e(k ) = yk − y k (23)
xk (5) = aDCα DC
X (k ) = SG N ( xk ) (24)
Where input state vector is given by: And

xk = [ xk (1) xk (2) xk (3) xk (4) xk (5) ] (15) S G N ( xi ) = + 1 if xi > 0 (25)


= − 1 if xi < 0
The measurement matrix is given by:
i = 1, 2 , ......, 2 N + 2
H k = [sin (ω kT s + φ 1 ) sin (3 ω kT s + φ 3 ) (16) The learning parameter α is chosen to lie between 0 < α < 2
sin (5 ω kT s + φ 5 ) 1 − ( kT s )]
for making the tracking error e( k ) converge asymptotically
The state transition matrix is given by: to zero. The learning parameter α can be made adaptive using
the following expression:
⎡1 0 0 0 0 ⎤
⎢0 1 0 0 0 ⎥⎥ α ( k ) = α 0 /(1 + k / β ) (26)
(17)
Fk = ⎢
⎢0 0 1 0 0 ⎥
⎢ ⎥ The estimated phases of different harmonics can be derived as:
⎣0 0 0 1 α DC ⎦
φ n = t a n − 1 (W 2 N / W 2 N − 1 ) (27)
The amplitudes can be estimated by taking the absolute values W1
of the state variables. X1

B) Phase Estimation Using Adaline: 978-


W2
The following equation gives the idea of using an adaptive X2
linear combiner comprising a neural network called ‘Adaline’
y^(k)
to estimate the phases of harmonics.
X3 W3
N
zk = ∑[an sin(ωn kTs )cosφn +an cos(ωn kTs )sin φn ] .
n=1 (18)
+aDC − aDCαDC kTs + vk .

Xn Wn e(k)
The input to the network is given by:

xk = [ a1 sin(ω1 kTs ) a1 cos(ω1 kTs ) . an sin(ω n kTs ) Weight


(19)
an cos(ω n kTs ) 1 − kTs ]T Updation
Fig.1 Block Diagram of Adaline
Also the associated weight vector of this network is given by:
IV. SIMULATION RESULTS
W ( k ) = [W 1 ( k ) W 2 ( k ) ....... W 2 N − 1 ( k ) (20) The computer simulations are carried out using Mat lab 7.1
W 2 N ( k ) W 2 N + 1 ( k ) W 2 N + 2 ( k )] T using the above two test signals. All the simulations are
performed in presence of Gaussian White Noise considering
Where the weight vector for the inputs given by 30 dB SNR. Simulation results for estimated fundamental,3rd
and 5th harmonic amplitudes and phases of two different test
W = [cos(φ1 ) sin(φ1 ) ....... cos(φN ) sin(φN ) aDC aDCαDC ] (21)

3
signals are shown to show the tracking and estimation
0.75
capability of the filter.
KAL
0.7 HINF
2.5
0.65
2

AMPLITUDE(P.U.)
0.6
1.5

1 0.55
AMPLITUDE(P.U.)

0.5
0.5

0
0.45
-0.5

0.4
-1 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
TIME IN SECOND
-1.5

-2
Fig.2(c) Estimated 5th Harmonic Amp
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
TIME IN SECOND

1.5
Fig.2 Test Signal 1 used for Estimation KALMAN
ADALINE

1.1
KAL
HINF PHASE(RADIAN) 1
1

0.9
AMPLITUDE(P.U.)

0.8 0.5

0.7

0.6
0
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
0.5 TIME IN SERCOND

0.4
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 Fig.2 (d) Estimated Fundamental Harmonic Phase
TIME IN SECOND

1.5
Fig.2 (a) Estimated Fundamental Harmonic Amp.
1.4 KALMAN
ADALINE
0.9 1.3

KALMAN 1.2
0.85 HINF
PHASE(RADIAN)

1.1
0.8
1
AMPLITUDE(P.U.)

0.75
0.9

0.7 0.8

0.65 0.7

0.6
0.6
0.5
0.55 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
TIME IN SECOND

0.5
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
TIME IN SECOND Fig.2 (e) Estimated 3rd Harmonic Phase

Fig.2 (b) Estimated 3rd Harmonic Amp.

4
1 0.58
ORG
0.8 HINF
0.56
KAL
0.6 KALMAN
0.54
ADALINE
0.4

AMPLITUDE(P.U.)
PHASE(RADIAN)

0.52
0.2
0.5
0

0.48
-0.2

-0.4 0.46

-0.6 0.44

-0.8
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 0.42
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
TIME IN SECOND
TIME IN SECOND

Fig.2 (d) Estimated 5th Harmonic Phase Fig.3 (b) Estimated 3rd Harmonic Amp

2
0.24

ORIGINAL
1.5 0.23
HINF
KAL
0.22
1
AMPLITUDE(P.U.)

AMPLITUDE(P.U.)

0.21
0.5
0.2

0 0.19

0.18
-0.5

0.17
-1
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
TIME IN SECOND 0.16
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
TIME IN SECOND
Fig.3 Test Signal 2 used for Estimation
Fig.3(c) Estimated 5th Harmonic Amp
0.7

ORGINAL 1.4
0.65
KALMAN
ADALINE
HINF 1.2
0.6 KALMAN

1
AMPLITUDE(P.U.)

0.55
PHASE (RADIAN)

0.8
0.5

0.6
0.45

0.4
0.4

0.2
0.35

0
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
TIME IN SECOND
-0.2
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
TIME IN SECOND
Fig.3 (a) Estimated Fundamental Harmonic Amp
Fig.3 (d) Estimated Fundamental Harmonic Phase

5
47, no. 10, pp. 2853–2856, Oct. 1999.
0.9

0.8
[3]. P. K. Dash, A. K. Pradhan, and G. Panda, “Frequency estimation of
distorted power system signals using extended complex Kalman filter,”
0.7 ADALINE IEEE Trans. Power Del., vol. 14, no. 3, pp. 761–766, Jul. 1999.
KALMAN
0.6
[4]. P. K. Dash, R. K. Panda, and G. Panda, “An extended complex
PHASE (RADIAN)

0.5 Kalman filter for frequency measurement of distorted signals,” IEEE


0.4
Trans Instrum. Meas., vol. 49, no. 4, pp. 746–753, Aug. 2000.

0.3
[5]. K. R. Shih and S. J. Huang, “Application of a robust algorithm for
0.2
dynamic state estimation of a power system,” IEEE Trans. Power Syst.,
0.1 vol. 17, no. 1, pp. 141–147, Feb. 2002.
0
[6]. M.Bettayeb, U.Qidwai, “A hybrid least squares-GA based algorithm
-0.1 for harmonic estimation,” IEEE Trans. Power Delivery, vol.18, no.2,
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
TIME IN SECOND Apr.2003

Fig.3 (e) Estimated 5th Harmonic Phase [7]. Chien-Hung Huang, Chien-Hsing Lee, Kuang-Jung Shih, and Yaw-
Juen Wang, “Frequency Estimation of Distorted Power System Signals
V. DISCUSSION Using a Robust Algorithm,” IEEE Trans. Power Syst., vol. 23, no. 1,
pp. 41–51, Jan. 2008.
Kalman Filter (KF) is designed for handling signals, which are
corrupted with only white noise where H-infinity filter is a [8]. M.Joorabian, S.S. Mortazavi, A.A. Khayyami, “Harmonic Estimation
robust estimator irrespective of types of noises. First H- in a Power System using a novel Hybrid Least Squares-Adaline
Algorithm” Elsevier Science Electric Power System Research
infinity filter is chosen for amplitude estimation because it is 79(2009) 107-116.
really efficient for tracking dynamic and abrupt changes of
harmonic amplitudes. Only two parameters i.e. model error
covariance Q and measurement error covariance R is tuned for [9]. P.K.Dash, D.P.Swain, A.Routray, A.C.Liew, Harmonic Estimation in a
accurate estimation. Second, the proposed Adaline algorithm power system using adaptive perceptrons, IEE Proc-Gener. Transm.
Distb. 143(November (6)) (1996)565-574
is a very simple and consequently very fast neural network,
which is composed of only one neuron, can estimate the
phases of harmonics using weight updation strategy.
Compared to KF algorithm [1, 3, 4, 5, 7] with equation of
proposed method H-infinity-Adaline needs less computational
efforts than Kalman Filter. So, for the same number of
harmonic components, the proposed algorithm will be faster.
Furthermore, although the selected area of application is
power systems, the same algorithm can be applied to other
types of signals from communication channels, telephones,
and other encrypted signals.
VI. CONCLUSION

A new hybrid H infinity-Adaline algorithm for estimation of


harmonic parameters in a power system is presented which is
essential to design a filter to reduce the effect of harmonics.
Both the nonlinear approaches effectively estimate the
amplitudes and phases of different harmonics and simulation
result reveals the improvement in the performance of the
proposed hybrid algorithm even in the presence of white noise
and decaying DC component.
REFERENCES

[1]. K. Nishiyama, “A nonlinear filter for estimating a sinusoidal signal


and its parameters in white noise: On the case of a single sinusoid,”
IEEE Trans. Signal Process., vol. 45, no. 4, pp. 970–981, Apr. 1997.

[2]. K. Nishiyama, “Robust Estimation of a Single Complex Sinusoid in


White Noise-H∞ Filtering Approach” IEEE Trans. Signal Process., vol.

You might also like