Suitable Mother Wavelet for Harmonics and

Interharmonics Measurements Using Wavelet Packet

Transform
Walid G. Morsi M. E. El-Hawary
Department of Electrical and Computer Engineering Department of Electrical and Computer Engineering
Sexton Campus, Dalhousie University Sexton Campus, Dalhousie University
Halifax, Canada Halifax, Canada
[email protected] [email protected]

Abstract—When considering harmonic distortion, the Fast Traditionally the Fast Fourier Transform (FFT) has been
Fourier Transform (FFT) performs well in calculating Root used to analyze harmonics in power system. The FFT is able
Mean Square (RMS) values of some distorted waveforms. to give accurate results only if the analyzed waveform fulfills
However, in the case of inter-harmonics the FFT produces large certain conditions [1]. Among these conditions, the signal
errors in estimating these RMS values. This paper introduces the
should not contain inter-harmonics otherwise, a spectral
use of Wavelet Packet Transform (WPT) for RMS measurements
in case of harmonics and inter-harmonics distortion. In order to leakage occurs and the FFT produces enormous errors.
minimize the spectral leakage associated with WPT the paper The wavelet transform has been introduced for harmonic
proposes an algorithm that is capable of choosing the most measurements of power systems [1]-[5]. The wavelet transform
suitable wavelet family and hence mother wavelet. This reduces is capable of accurately extracting the information contained in
the error associated with wavelet analysis. The proposed any distorted waveform with high resolution both in time and
algorithm is applied to two case studies considering harmonics frequency. For harmonic measurement purposes, the frequency
and inter-harmonics distortion and the results are compared with spectrum should be divided into uniform bands. This can only
the true values and the FFT based values. be realized using the Wavelet Packet Transform (WPT). The
number of frequency bands is related to the number of
Keywords-component; harmonics; inter-harmonics; power decomposition levels which depend on the harmonic order that
quality; wavelets. is required to be measured. The choice of the wavelet family
and the mother wavelet greatly affects the accuracy of the
I. INTRODUCTION measurements [6]. In this paper an algorithm is proposed to
choose the most suitable wavelet family and mother wavelet
Electric power quality (EPQ) has gained much attention
for measuring Root Mean Square (RMS) values of distorted
during the past few years because of the increased use of non- waveforms using Wavelet Packet Transform. The paper
linear loads that inject distorted currents into the power system considers the cases of harmonics and inter-harmonics through
networks and hence distorting the supply voltage nodes the application of the proposed algorithm to two case studies.
through generating distorted voltage drops. Non-linear loads The organization of this paper is as follows: Section II briefly
such as Personal Computers (PCs), inverters and power explains the basics of the Wavelet Packet Transform, while
electronic equipment are mainly responsible for injecting section III fully explains the proposed algorithm for choosing
harmonics to the power system. Harmonics have harmful the most suitable wavelet family and mother wavelet. Section
effects such as excessive heating of power system IV contains two case studies considering harmonics and inter-
components, malfunction of equipments, mis-operation of harmonics, also it shows the results of the application of the
protection schemes, and introduction of errors to the proposed approach and comparisons with true values and FFT
measurement devices such as ammeters, voltmeters and watt- based values are also included. Finally conclusions are given in
meters. Sometimes inter-harmonics, those are non-integer section V.
multiples of the fundamental frequency, appear in the power
system waveforms. Inter-harmonics are the result of using II. WAVELET PACKET TRANSFORM
remote controls, adjustable speed drives, back-to-back Wavelet Transform is a windowing technique to represent
converters operating at different frequencies, asynchronous any distorted waveform into a time-frequency domain. Thus
switching of semiconductor devices in static converters such decomposing the waveform into “approximations” that contain
as cycloconverters and pulse width modulation (PWM) low frequency components and “details” that contain the high
converters. These inter-harmonics may be located anywhere in frequency components at every time sample. When successive
the frequency spectrum of the power system waveforms and decompositions are performed on the approximations and the
their effect is increasing the voltage and current distortion, details the Wavelet Packet Transform (WPT) can be obtained
increasing the heat loss and increasing the flicker level.

748
0840-7789/07/$25.00 ©2007 IEEE
with uniform frequency bands. Fig. 1 shows the WPT refers to node number such that n ∈ N , ( j , k ) ∈ Z .
2
decomposition tree with three levels of decomposition.
Therefore for every k the set function W j , n can be computed
using (3) producing the (j,n) wavelet packet shown in Fig. 1.

(a) (b)

Figure 1. Wavelet Packet decomposition tree.

From Fig. 1 at each decomposition level there are nodes

(c) (d)
which represent the components of the original waveform at
that level of decomposition and at different frequencies. The Figure 2. Wavelet families. (a) db10, (b) sym 8, (c) coif 5 and (d) dmey.
terminal nodes of the tree are called leaves and they represent
the different components at different frequencies constituting
the original waveform. The wavelet packets can be generated 2 2
by considering a sequence of functions
( Wn ( x), n = 0,1, 2, 3,..... ) as follows: 0 0

2N − 1
W
2n
(x) = ∑ h ( k ) Wn ( 2 x − k )
2 (1) -2
0 100 200
-2
0 100 200
k=0 W0 W1
2N − 1 2 2
W
2n + 1
( x ) = 2 ∑ g (k ) W (2 x − k )
n
(2 )
k =0 0 0
Here, W0 ( x ) , W1 ( x) are the scaling and wavelet
-2 -2
function, h(k ) , g (k ) are two filters of length N respectively. 0 100 200 0 100 200
Different wavelet functions exist, and functions with common W2 W3
characteristics are grouped together into families. The function 2 2
that is associated with the approximations is called the scaling
function while those associated with the details are called 0 0
wavelet functions. The complete history and citation of the
characteristics of each family can be found in [7]. Four -2 -2
0 100 200 0 100 200
families were used here; Daubechies (db), Symlets (sym), W4 W5
Coiflets (coif) and Discrete Meyer (dmey). See Fig. 2.
2 2
Note that the number beside the wavelet name is its number
of vanishing moments. Thus applying (1) and (2) for the dmey
for n = 0 to 7 which corresponds to three levels of 0 0
decomposition we obtain the following W-functions shown in
Fig. 3. Having obtained the W-functions, the analyzing -2 -2
functions can be obtained from: 0 100 200 0 100 200
W6 W7
− j/2 − j
W
j , n, k
( x) = 2 W (2
n
x − k) (3 )
Where, j is the scale or level index, k is the time index and n Figure 3. Discrete Meyer Wavelet Packet.

749
 III. SUITABLE FAMILY AND MOTHER WAVELET SELECTION
ALGORITHM
As mentioned previously, the accuracy of the Wavelet
Transform depends on the choice of the wavelet family along
with the mother wavelet (i.e., wavelet with suitable number of
vanishing moments). There is a need to choose the best family
and mother wavelet that give best and accurate results. The
algorithm proposed here consists of the following steps:
1) Start with any arbitrary family and any arbitrary (a)
mother wavelet. (For example start with Daubechies
family and db1 as mother wavelet).
2) Apply WPT to the sampled distorted waveform.
3) Reconstruct the coefficients at the terminal nodes (tree
leaves).
4) Calculate the RMS values X at each terminal node
using the following relation [8]:
N− j j
1 2 −12 −1
X
j, n, k
= ∑ ∑ ( d nj , k ) 2 (4 )
2N n=0 k =0
(b)
n
where d j ,k is the wavelet coefficient at level j, time k Figure 4. Time domain and FFT for the voltage waveform of the first case
study.
and node n.
5) Calculate the error which is the difference between the
estimated values (true values) and the calculated values
2.5
at each terminal node.

.
6) Calculate the Root Mean Square of the errors obtained 2
from step 5.
1.5
7) Repeat steps 2 to 6 for the remaining mother wavelet
belonging to this family.
RMS Error

1
8) Repeat steps 2 to 7 for the remaining families.
0.5
Then the family and the mother wavelet with the least
values of root mean square errors is the most suitable one. 0
This way the problem of spectral leakage associated with db10 sym8 coif5 dmey
wavelet analysis can be reduced.
Best mother wavelet
IV. CASE STUDIES AND RESULTS

A. First Case Study: Harmonic Distortion Figure 5. Best mother wavelet for each family, first case study.
Fig. 4 shows a distorted voltage waveform and its FFT
spectrum. The waveform contains the fundamental (60Hz) with Table I. True RMS values, FFT based RMS values and WPT based RMS
values for first case study
100 volts amplitude, third component with 35 volts amplitude,
fifth component with 15 volts amplitude and seventh harmonic
True RMS FFT Based WPT Based
with 10 volts amplitude. The sampling frequency is taken as Frequency
values RMS values RMS values
1.92 KHz, and the window size of the FFT is 0.0167 sec. After
60 70.7106 70.7106 70.6950
applying the algorithm explained earlier the best mother
wavelet of each family are identified with their corresponding 180 24.7487 24.7487 24.3367
root mean square errors. Fig. 5 shows the best mother wavelet 300 10.6066 10.6066 10.4572
of each family and the associated root mean square error. 420 7.0710 7.0710 7.2745
It is clear from Fig. 5, that the best wavelet family and 540 0.000 0.000 2.0891
mother wavelet is the (dmey) when using Wavelet Packet 660 0.000 0.000 0.5509
Transform, Also the figure shows that sym8 has the highest
780 0.000 0.000 0.2510
error and therefore it is not suitable for this analysis. Table I
lists the true RMS values, FFT based RMS values and WPT 900 0.000 0.000 0.0982
based RMS values.

750
B. Second Case Study: Harmonic and Inter-harmonic lists the true RMS values, FFT based RMS values and WPT
Distortion based RMS values.
In this case the voltage contains in addition to the Table II. True RMS values, FFT based RMS values and WPT based RMS
fundamental (60Hz) with 100 volts amplitude, the fifth values for second case study
harmonic component with 50 volts amplitude and an inter-
harmonic component at 3.25 of the fundamental, which is True RMS FFT Based WPT Based
Frequency
clearly non-integer multiple of the fundamental frequency values RMS values RMS values
component with 50 volts amplitude. Fig. 6 shows the time 60 70.7106 71.9700 70.6816
domain waveform and its FFT spectrum. Fig. 7 shows the root 180 35.3553 31.050 34.5608
mean square error associated with each best mother wavelet of 300 35.3553 31.870 35.5538
each family.
420 0.000 2.370 3.1686
540 0.000 2.070 1.2086
660 0.000 1.730 0.6446
780 0.000 1.550 0.2542
900 0.000 1.480 0.1034

From Table II, the FFT produces enormous errors and the
RMS values based on FFT are far from the true values, on the
other hand the WPT based RMS values are close to the true
(a) values in almost all the levels. At the fundamental frequency
the FFT method produces a percentage error of 1.78 %
however the WPT method percentage error is -0.041%. For
inter-harmonic estimation, the FFT produce percentage error of
-12.177% while the WPT percentage error is only -2.247%. At
the fifth harmonic the FFT percentage error is -9.858% while
that of WPT method is 0.56144%. Therefore it is clear that in
case of inter-harmonics the FFT introduces large error not only
for the estimation of the inter-harmonic components but also in
the other components even that are integer multiple of the
fundamental frequency. On the other hand the RMS values
based on the WPT method are very close to the true values for
inter-harmonics and also the values of other harmonics that
exists in the spectrum are not affected unlike the case for FFT.
(b)
Figure 6. Time domain and FFT for the voltage waveform of the second case V. CONCLUSIONS
study.
This paper presents an algorithm for selecting the most
suitable wavelet family and mother wavelet when measuring
1.8
RMS values of harmonics and inter-harmonics using the
Wavelet Packet Transform (WPT). The proposed algorithm has
1.6
been tested by applying it to two case studies. The first case
.

1.4 study considers harmonic distortion at different frequencies

1.2 while the second case study considers the presence of inter-
1 harmonics along with harmonic distortion. The results indicate
0.8 that the proposed algorithm is powerful, accurate and greatly
RMS Error

0.6 reduce the problem of spectral leakage that is associated with

0.4 Wavelet Transform. Also the results obtained show that the
0.2 FFT performs well for harmonic distortion while gives
0
enormous errors in case of inter-harmonics. On the other hand
WPT gave good results with very small error in case of
db10 sym8 coif5 dmey
harmonic distortion but in case of inter-harmonics give values
Best mother wavelet that are very close to the true values and the errors are
significantly smaller than those of FFT. Moreover calculating
the RMS values using the WPT is more beneficial since the
Figure 7. Best mother wavelet for each family, second case study. WPT can preserve both time and frequency information with
high resolution by providing time-frequency spectrum, while
From Fig.7 the best mother wavelet is again (dmey) on the
other hand (coif5) gives large error values in this case. Table II

751
FFT can only provide an amplitude frequency spectrum hence [4] G. Takata, J. Tahara, M. Michihira, A. Tsuyoshi, K. Amako, H. Omori
no information about time can be obtained. and K. Yasui, “The time-frequency analysis of the harmonics with
wavelet transform for the power electronics systems,” in Proc. the 2002
Power Conversion Conference PCC Osaka 2002, vol. 2, pp. 733-737,
REFERENCES April 2002.
[1] J. Driesen, T. Van Craenenbrock, R. Reekmans and D. Van Dommelen, [5] J. Barros and R. I. Diego, “Application of the wavelet packet transform
“Analysizing time-varying power system harmonics using wavelet to the estimation of harmonic groups in current and voltage waveforms,”
transform,” in Proc. 1996 IMTC-96 Instrumentation and Measurement IEEE Trans. Power Delivery, no. 1, vol. 21, pp. 533-535, January 2006.
Technology Conference, vol. 1, pp. 474-479, June 1996. [6] A. Domijan, A. Hari and T. Lin, “On the selection of appropriate filter
[2] L. Eren and M. J. Devaney, “Calculation of power system harmonics via bank for power quality monitoring,” Int. J. Power Energy Syst.
wavelet packet decomposition in real time metering,” in Proc. 19th (PowerCon Special Issue), pp. 46-50, 2004.
Instrumentation and Measurement Technology Conference IMTC/2000, [7] M. Misiti, Y. Misiti, G. Oppenheim and J. M. Poggi, “Wavelet toolbox
vol. 2, pp. 21-23, May 2000. user’s guide for use with MATLAB,” The mathworks, Inc. 2006.
[3] K. P. Wong and V. L. Pham, “Analyzing power system waveforms using [8] E. Hamid, Z. Kawasaki and R. Mardiana, “Wavelet packet transform for
wavelet packet transform approach,” in Proc. 5th International RMS and power measurements,” in Proc. 2001 IEEE Power Engineering
Conference on Advances in Power System Control, Operation and Society Summer Meeting, pp. 1243-1245, July 2001.
Management, APSCOM2000, vol. 2, pp. 500-504, October 2000.

752