Professional Documents
Culture Documents
Total Harmonics Distortion Reduction Using A New Method of Adaptive Filtering
Total Harmonics Distortion Reduction Using A New Method of Adaptive Filtering
Abstract—This paper presents a novel method to reduce the This paper focuses on the single phase inverter which is shown
Total Harmonics Distortion (THD) using adaptive filters. Least in Fig. 1. The inverter shown in Fig.1 consists of four choppers
Mean Square (LMS) and Normalized Lest Mean Square (NLMS) (Insulated-Gate Bipolar Transistor (IGBT) transistors with
were used as adaptation algorithms. Both of these algorithms diodes in parallel). In order to control the turning on and off of
gave a very good performance in terms of distortion reduction. the IGBT circuits, different kinds of controllers can be used.
Instead of focusing on specific harmonics as in other methods of The controllers command the switches such that when IGBT1
adaptive filtering, this method targets all the harmonics is on, IGBT2 will be off and vice versa [1]. Pulse Width
contained in the signal for elimination. A single phase inverter Modulation (PWM) is the most common method for
was used as a study case to test the performance of the proposed
controlling the switches of the inverters [1].
technique. The obtained results showed that the presented
method can decrease the total harmonics distortion dramatically
over time.
I. INTRODUCTION
In recent years, there has been growing and wide spread use
of power electronics inverters. The main goal of inverters is to
change direct current DC to alternating current AC with the
desired amplitude and frequency. The output current waveform
of an ideal inverter is a pure sinusoidal signal but the common
output signals usually contain distortions caused by harmonics
[1]. Besides the fundamental frequency f0, the output signal Fig. 1: Single phase DC-AC converter.
will contain harmonics at frequencies of 2f0, 3f0, 4f0, and so
on. The odd harmonics are the most dominant ones. To reduce III. ADAPTIVE FILTERS
the effect of these harmonics several studies have been done
Adaptive filters are linear filters with the ability to change
using different techniques; active power filters as in the work
their weights iteratively in time in order to get the optimum
of Ilakkia and Rajalakshmi [2], Rahmani et al [3] and Dixon et
solution for the cost function. Fig. 2 illustrates the main
al [4], passive filters such as the work of Rivas et al [5],
concept of adaptive filter. The generated reference signal is the
artificial intelligent (AI) as in the work of Eyad and Asumadu
input to the adaptive filter. The output of the filter will be
[6]. Besides active filters, passive filters, and AI, several
subtracted from the input signal in order to compute the error
studies have used adaptive filters. Almost all of the adaptive
signal, which is controlling the filter weights to reduce the
filtering studies have focused on specific harmonics such as the
error as much as possible. Based on the definition of the cost
work of Blasko et al [7]- [8], and Rode and Ladhake [9]- [10].
function and the known and unknown system parameters, there
The majority focused on the third and/or fifth harmonics only;
are many ways to reach the optimization goal. For full
known as adaptive selective harmonics elimination (ASHE).
understanding of adaptive filtering, [11] and [12] are good
This paper proposes the use of adaptive filters to reduce all the
references to look at. Many studies have been presented
harmonics in the distorted signal. The rest of the paper is
recently on using adaptive filters for harmonic reduction in
organized as follow: Section II illustrates the main concept of
power and power electronics devices. The majority of them
DC to AC converters, Section III will explain the basic concept
have focused on one or two harmonic components as shown in
of adaptive filters, Section IV will discuss the simulation and
Fig. 3. In this paper, two algorithms of adaptive filtering were
results of the proposed model, and finally Section V is the
used; LMS and NLMS. In both algorithms, the cost function is
conclusions and future works.
defined as the mean square error between the estimated output
and the actual output. The same concept can be generalized to
II. DC-AC CONVERTERS other algorithms as well.
DC to AC converters (inverters) have been used in many
applications such as renewable energy, transportation, and
induction heating. There are three types of inverters: single
phase inverters, three phase inverters, and multi-level inverters.
where T refers to the transpose of the matrix and w(n) is the
weight vector which is:
w(n) [ w(0) w(1) . . . w( M 1)]T (3)
The main idea of LMS and other adaptive algorithms is to
use steepest decent method to find the filter weight, which
minimize the cost function. This process started by defining the
cost function C(n) as the expected value of the square of the
error signal e(n). The error signal is the difference between the
desired output d(n) and the actual one y(n):
Fig. 2: Adaptive Filters.
e(n) [e(n) e(n 1) . . . e(n M 1)]T (4)
d (n) [d (n) d (n 1) . . . d (n M 1)]T (5)
Based on the definition of the cost function which is:
2
C (n) E e (n) (6)
where E [] is the expected value and using steepest decent
method to solve (6) means taking the partial derivatives with
respect to the individual filter weights:
2 e( n )
w C ( n ) w e (n) 2 e(n) w
E (7 )
But e(n) d (n) x(n) w(n) T (8)
e( n)
Hence x(n) (9)
w
To take the minimum of the cost function, we need to take
a step in the opposite direction of w C (n ) . To express this
point mathematically:
2 w
Fig. 3: ASHE block diagram. w( n 1) w( n) C ( n) (10)
A. LMS
w( n 1) w( n) . 2. e( n) . x( n) (11)
LMS is the most common algorithm that has been used to 2
reduce harmonics in different applications. Selecting the right w(n 1) w(n) . x(n) . e(n) 12
value of step size is very important to control the stability and
convergence speed of LMS algorithm. In this algorithm, the where is the step size. Equation (12) is the weight update
cost function is defined as the mean square error between the equation which is the relationship between the current weight
estimated output and the actual one. The weights, error, and vector and the next one.
output equation of LMS algorithm can be derived as follow
B. NLMS
[11]- [12]:
The filter length and the power of the signal affect the
For a given adaptive filtering system with input matrix stability, convergence, and steady state behavior of the LMS
algorithm. To solve this problem, NLMS which is a variant of
x(n) which is M*M matrix: the LMS, has been used widely. It solves the problem by
normalizing the power of the input. In NLMS algorithm, all the
x(n) [ x(n) x(n 1) . . . x(n M 1)] (1) equations mentioned in the previous section stay the same
except the weight update equation will change as follow [11]-
[12]:
where M is the filter taps and n is the current input samples.
The output of the adaptive filter will be:
y(n) [ y (n) . . . y(n M 1]T x(n) w(n)T (2)
w(n 1) w( n) 2
* x ( n)* e( n) (13)
x ( n)
2
where x(n) x ( n) H * x ( n ) is the normalizing factor, H
400
200
distorted
-200
-400
400
200
distorted as shown in Fig. 5. The current signal was used in this -200
-400
study. LMS and NLMS algorithms were applied to get clean 0.1 0.105 0.11 0.115 0.12 0.125 0.13 0.135 0.14 0.145
0
described in Section III for both algorithms. Instead of
-200
focusing on specific harmonics as in ASHE, all harmonics
-400
contained in the distorted signal were targeted for elimination. 0.005 0.01 0.015 0.02 0.025 0.03 0.035 0.04 0.045 0.05
LMS and NLMS algorithms gave a very clean signal compared
to the distorted one as shown in Figs. 7 - 11. The use of
400
adaptive filters can reduce the THD from 5% to 2.6 - 3%,
200
approximately. Both algorithms are very powerful in reducing
cleaned
different from the rated values but there is no overshoot in the -400
0.005 0.01 0.015 0.02 0.025 0.03 0.035 0.04 0.045 0.05
signal. This fluctuation in the early stage gives a good
explanation of the high THD value in the signal even of it
looks very clean. In LMS algorithm, the step size has to be Fig. 8: Distorted and clean current signal using LMS algorithm when µ=1e-8
very small (example 1e-8 range) to reach the adaption goal and M=120 showing the early adaptation steps.
while in NLMS the goal is reached with higher values of step
size (example 0.01-0.01) as shown in Figs. 7 - 11.
Instead of eliminating single specific harmonic as in ASHE, all
the harmonics in the signal were targeted for elimination. LMS
400
200
-200
-400
Both algorithms gave very good performance in term of
0.1 0.105 0.11 0.115 0.12 0.125 0.13 0.135 0.14 0.145 0.15
distortion reduction. LMS used a very small value of step size
to reach the adaptation goal while NLMS can reach this goal
even with high values of step size. The THD was reduced by
400
200
more the 50% from its previous values using the new proposed
cleaned
-200
method. As a future work, more investigation will be done on
how to reduce the adaptation time required for each algorithm
-400
0.1 0.105 0.11 0.115 0.12 0.125 0.13 0.135 0.14 0.145 0.15
0
[2] A. Ilakkia and R. Rajalakshmi, “Selective harmonic elimination using
three phase shunt active filter,” International Journal of Advanced
-200
-400
0.1 0.105 0.11 0.115 0.12 0.125 0.13 0.135 0.14 0.145 Research in Electrical, Electronics and Instrumentational Engineering,
vol. 3, no. 3, pp. 7648-7653,March, 2014.
400
[3] S. Rahmani, A. Hamadi and K. Al-Haddad, “A new combination of
200
shunt hybrid power filter and thyristor controlled reactor for harmonics
cleaned
-200
-400
Power & Energy Conference (EPEC),pp.1-6, 2009.
0.1 0.105 0.11 0.115 0.12 0.125 0.13 0.135 0.14 0.145 0.15
[4] J. W. Dixon, G. Venegas, and L. Morán, “A series active power filter
based on a sinusoidal current controlled voltage-source inverter,” IEEE
Fig. 10: Distorted and clean current signal using NLMS algorithm when µ=0.1 Transactions on Industrial Electronics, vol.44, no. 5, pp.612-620, 1997.
and M=120. [5] D. Rivas, L. Moran, J. W. Dixon and J. R. Espinoza, “Improving passive
filter compensation performance with active techniques,” IEEE
Transactions on Industrial Electronics, vol. 50, pp. 161- 170, 2003.
[6] E. Almaita and J. Asumadu, “Radial basis function neural
400 networks(RCFNN) and p-q power theory based harmonics identification
200 in converter,” Journal of Power Electronics, vol.11, pp. 922-930,
distorted
0
November, 2011.
-200
-400
[7] V. Blasko, L. Arnedo, P. Kashirsagar, and S. Dwari, “Control and
0.005 0.01 0.015 0.02 0.025 0.03 0.035 0.04 0.045 0.05 elimination of sinusoidal harmonics in power electronics equipment: A
system approach,” IEEE Energy Conversion Congress and Exposition,
400
pp. 2827-2837, 2011.
200
[8] V. Blasko, “A novel method for selective harmonic elimination in
cleaned
0
power electronic equipment,” IEEE Transaction on Power Electronics,
vol. 22,no.1, pp. 223-228,January, 2007.
-200
-400