a new classification for power quality events in distribution systems
TRANSCRIPT
A new classification for power quality events in distribution systems
Okan Ozgonenela , , ,
Turgay Yalcina , ,
Irfan Guneyb , ,
Unal Kurtc ,
a Ondokuz Mayis University, Electrical & Electronic Eng. Depart., Kurupelit, Samsun, Turkey
b Acibadem University, Dean of Engineering Faculty, Gulsuyu Mah. Fevzi Cakmak Cad. No. 1, Maltepe, Istanbul, Turkey
c Amasya University, Technology Faculty, Electrical & Electronic Eng. Depart., Amasya, Turkey
Received 19 January 2012
Revised 22 July 2012
Accepted 18 September 2012
Available online 12 October 2012
http://dx.doi.org/10.1016/j.epsr.2012.09.007, How to Cite or Link Using DOI
Permissions & Reprints
Abstract
This paper presents the performance evaluation of support vector machine (SVM) with one against all (OAA) and different classification methods for power quality monitoring. The first aim of this
study is to investigate EEMD (ensemble empirical mode decomposition) performance and to compare it with classical EMD (empirical mode decomposition) for feature vector extraction and selection of power quality disturbances. Feature vectors are extracted from the sampled power
signals with the Hilbert Huang Transform (HHT) technique. HHT is a combination of EEMD and Hilbert transform (HT). The outputs of HHT are intrinsic mode functions (IMFs), instantaneous
frequency (IF), and instantaneous amplitude (IA). Characteristic features are obtained from first IMFs, IF, and IA. The ten features—i.e., the mean, standard deviation, singular values, maxima and
minima—of both IF and IA are then calculated. These features are normalized along with the inputs of SVM and other classifiers.
Highlights
► Hilbert Huang Transform is first introduced for all possible power quality issues. ► Feature extracting and selection are done by an effective way. ► Singular value decomposition method is
applied to get additional features. ► Support vector machine is used as classifier of all possible power quality issues. ► Performance of the suggested algorithm is then compared to well-known
algorithms.
Keywords
Power quality disturbances ;
Hilbert Huang Transform ;
Ensemble Empirical Mode Decomposition ;
Support Vector Machines ;
One-against-all method
1 .Introduction
Power monitoring has taken on an important role in power systems. Power quality (PQ); or, more specifically, a power quality disturbance, is generally defined as any change in power (voltage,
current, or frequency) that interferes with the normal operation of electrical equipment. Power quality concerns various kinds of electric disturbances such as voltage sag/swell, flicker, impulse,
harmonics, DC component of voltage, oscillatory transients, electromagnetic interference (EMI), and power system outage [1] and [2]. However, more generally, PQ disturbances can be classified as in three groups. These are Frequency and amplitude variations and transient phenomena. Amplitude
variations cover voltage variations under normal operating conditions, excluding faults and interruptions, short and long interruptions, voltage sags (dips) and swells, and voltage fluctuations
known as flicker. Transient phenomena cover voltage unbalance and voltage/current waveform disturbances [3].
To analyze these disturbances, data are often available as a form of sampled time function that is represented by a time series of amplitudes. When dealing with such data, the Fourier transform (FT)-
based approach is most often used. FT assumes periodicity of a given signal and loses the time axis account; however, it is not capable of providing time information about signal disturbances. Short
time Fourier transform (STFT) provides both time and frequency information, but it suffers severely from the Heisenberg uncertainty principle [4], causing it to undergo a “trade-off” between time
resolution and frequency resolution. This is one of the reasons for the creation of the wavelet transform (or multiresolution analysis in general), which can give good time resolution for high-frequency events, and good frequency resolution for low-frequency events, which is the type of
analysis best suited for many real signals. Wavelet transform [5], which is a popular signal analysis method, offers continuous and discrete wavelet transforms (CWT and DWT) [6] and wavelet packet
transform (WPT) [7] and [8] for the feature extraction of signals.
Many classification algorithms have been developed by researchers for classification of power electrical disturbances. Integrated Fourier linear combiner and fuzzy expert systems [9] were used
for the classification of transient disturbance waveforms in power system. S-Transform and two dimensional time–time (TT) transform [10] have been implemented for electrical fault identification.
In [10], transient disturbance waveforms in power system were investigated by S-transform with together TT transform methods. It is reported that ST with together TT transform approaches was able to identify disturbed waveforms. An adaptive neural network approach for the estimation of
harmonic distortions and power quality in power networks was also implemented [11]. A hybrid system to automatically detect, locate, and classify disturbances affecting power quality in an electrical power system is also presented [12]. The least absolute value (LAV) state estimation
algorithm has been used to measure the flicker voltage magnitude [13], while the Simulated Annealing (SA) optimization algorithm has been used for measuring the voltage flicker magnitude,
frequency and the harmonics contents of the voltage signal for power quality analysis [14]. An algorithm to detect the fundamental frequency is proposed. It is based on the chirp-z transform
(CZT) spectral analysis and is able to observe all standards in force because of its accuracy and working characteristics [15]. Wavelet multi-resolution decomposition, which combines frequency
domain with time-domain analysis for power disturbance feature extraction, is also proposed [16]; moreover, a wavelet norm entropy-based effective feature extraction method for power quality
disturbance classification problem has been studied by [17]. Multi-wavelet based neural networks with learning vector quantization networks are used for power quality disturbances as a powerful
classifier [18]. Fuzzy ARTMAP, back propagation algorithm (BPA) and Radial Basis Function (RBF) networks, in combination with S Transform, Wavelet transform, and Hilbert Transform (HT) for
classifying power faults, have also been used [19].
On the contrary, unlike many of the aforementioned decomposition methods, EMD is intuitive and direct, with the basic functions based on and derived from the data. The assumptions for this
method are: (a) the signal has at least one pair of extrema; (b) the characteristic time scale is defined by the time between the successive extrema; and (c) if there are no extrema, and only inflection
points, then the signal can be differentiated to realize the extrema, whose IMFs can be extracted. Integration may be employed for reconstruction. The time between the successive extrema was
used by Huang et al. [20] as it allowed the decomposition of signals that were all positive, all negative, or both. This implied that the data did not have to have a zero mean. This also allowed a
finer resolution of the oscillatory modes.
This paper presents a hybrid power quality monitoring approach based on EEMD and SVM. EEMD and basic statistical calculations of its outputs are used for feature extraction. Moreover, SVM is used for classifying the analyzed power disturbances. To show the performance of the proposed
methods, different well-known classifiers are tested. Among all the tested classification algorithms, the suggested analytical procedures, followed by SVM-OAA, give a reliable solution for classifying
non-stationary power disturbances. The proposed PQ monitoring approach is easy to implement and presents reliable solution with high accuracy.
2 .Materials and methods
The suggested classification process is shown in Fig. 1.
<img class="figure large" border="0" alt="Full-size image (60 K)" src="http://ars.els-cdn.com/content/image/1-s2.0-S0378779612002891-gr1.jpg" data-
thumbsrc="http://ars.els-cdn.com/content/image/1-s2.0-S0378779612002891-gr1.sml" data-fullsrc="http://ars.els-cdn.com/content/image/1-s2.0-S0378779612002891-gr1.jpg>"
Fig. 1. Basic stages of the classification process.
View thumbnail images
Figure options
View in workspace
Download full-size image
Download as PowerPoint slide
As seen in Fig. 1, all major power quality issues such as voltage sag/swell, flicker, harmonics, transients, DC offset, electromagnetic interference (EMI), and interruptions have been investigated. Training procedure has been achieved by using synthetically generated data. Moreover, the supply frequency and flicker frequency are randomly chosen between 49.5 Hz and 50.5 Hz and between 8
Hz and 25 Hz, respectively. Sampling frequency is set at 25.6 kHz in compliance with the real-time data. This yields a 3200 × 3 data matrix for a three-phase system. The testing procedure was
completed by using real-time data supplied by TUBITAK (The Scientific and Technological Research Council of Turkey). The data collected by the National Power Quality Monitors (NPQMC) during an event breaking normal operating conditions are transferred by means of a network infrastructure
which is formed in the NPQMC. The data collected in the center are kept in databases and file systems after processing them on time and location basis to consider their distribution in the
country.
EEMD is mainly a signal processing technique to extract distinctive features; namely, IMFs. Feature selection requires a series of calculations based on statistics such as maxima, minima, singular value,
standard deviation, and mean. Next, the singular value decomposition (SVD) technique is used for calculating some additional properties of the features’ singular values. SVD was chosen simply
because the data matrix is not square. It has been observed that additional properties of the feature vectors increase the performance of the classification algorithm. Finally, SVM with a one-against-all
(OAA) approach is used for classification of PQ events.
Sections 2.1, 2.2, 2.3 and 2.4 give the mathematical background of the proposed feature extraction and selection algorithms.
2.1 .EMD
EMD is a self-adaptive signal-processing method that has been successfully applied in non-stationary signal processing. The EMD method is based upon the local characteristic time scale of the signal and
could decompose the complicated signal function into a number of IMFs. IMFs reflect the intrinsic and real physical information of analyzed the signals. Moreover, the superposition of IMF
components can reconstruct the analyzed signal (Eq. (1)). EMD method is a sifting process; to decompose one IMF component requires many sifting cycles. The sifting process actually serves two
purposes: to eliminate riding waves and to make the wave profiles more symmetrical with respect to zero (Fig. 2).
<img class="figure large" border="0" alt="Full-size image (46 K)" src="http://ars.els-cdn.com/content/image/1-s2.0-
S0378779612002891-gr2.jpg" data-thumbsrc="http://ars.els-cdn.com/content/image/1-s2.0-S0378779612002891-gr2.sml" data-fullsrc="http://ars.els-cdn.com/content/image/1-s2.0-
S0378779612002891-gr2.jpg>"
Fig. 2. A flowchart of EMD algorithm.
View thumbnail images
Figure options
View in workspace
Download full-size image
Download as PowerPoint slide
The EMD algorithm used in this study comprises the following steps:
(i)
Identify all the extrema (maxima and minima) of the signal, x(t).
(ii)
Generate the upper and lower envelope by the cubic spline interpolation of the extrema points developed in step (i).
(iii)
Calculate the mean function of the upper and lower envelope, m(t).
(iv)
Calculate the difference signal d(t) = x(t) − m(t).
(v)
If d(t) becomes a zero-mean process, then the iteration stop and d(t) is an IMF1, named c1(t); otherwise, go to step (i) and replace x(t) with d(t).
(vi)
Calculate the residue signal r(t) = x(t) − c1(t).
(vii)
Repeat the procedure from steps (i) to (vi) to obtain IMF2, named c2(t). To obtain cn(t), continue steps (i)–(vi) after n iterations. The process is stopped when the final residual signal r(t) is obtained
as a monotonic function.
At the end of the procedure, we have a residue r(t) and a collection of n IMF, named from c1(t) to cn(t). Now, the original signal can be represented as:
(1)
<img height="46" border="0" style="vertical-align:bottom" width="141" alt="View the MathML source" title="View the MathML source"
src="http://ars.els-cdn.com/content/image/1-s2.0-S0378779612002891-si1.gif>"
Often, r(t) is regarded as cn+1(t) [20].
The sifting process should be applied with care, for carrying the process to an extreme could make the resulting IMF a pure frequency modulated signal of constant amplitude. To guarantee that the IMF components maintain enough physical sense of both amplitude and frequency modulations, a criterion should be defined for the sifting process to stop. This can be accomplished by limiting size of the standard deviation (SD), it is similar to the Cauchy convergence test, computed from the two
consecutive sifting results as
(2)
<img height="49" border="0" style="vertical-align:bottom" width="198" alt="View the MathML source" title="View the MathML source" src="http://ars.els-
cdn.com/content/image/1-s2.0-S0378779612002891-si2.gif>"
where di(t) is proto (candidate)-mode function.
A typical value for SD can be set between 0.2 and 0.3. As a comparison the two Fourier spectra, computed by shifting only five out of 1024 points from the same data, can have an equivalent SD of 0.2–0.3 calculated point-by-point. Therefore, SD value of 0.2–0.3 for the sifting procedure is a very
rigorous limitation for the difference between siftings [20].
In Fig. 2, Extrema <3 indicates that the residue becomes a monotonic function from which no more IMFs can be extracted.
2.2 .EEMD
One of the major drawbacks of the EMD algorithm is the appearance of mode mixing. Therefore, the EEMD algorithm was introduced. The algorithm defines the IMF set for an ensemble of trials, each
one obtained by applying EMD to the signal of interest with added independent identically distributed white noise of the same standard deviation. The problem of mode mixing can be
overcome by taking into account the properties of white noise [21].
The steps for the EEMD algorithm are as follows:
i.
Add a white noise series n(t) to the targeted signal, named x1(t) in the following description, and x2(t) = x1(t) + n(t).
ii.
Decompose the data x2(t) with the EMD algorithm.
iii.
Repeat Steps (i) and (ii) until the trial numbers, regardless of added white noise, consistently register the same power. The new IMF combination Cij(t) is achieved when i is the iteration number and j is
the IMF scale.
iv.
Estimate the mean (ensemble) of the final IMF of the decompositions as the desired output:
(3)
<img height="46" border="0" style="vertical-align:bottom" width="160" alt="View the MathML source" title="View the MathML source"
src="http://ars.els-cdn.com/content/image/1-s2.0-S0378779612002891-si3.gif>"
where ni denotes the trial (ensemble) numbers.
(4)
<img height="36" border="0" style="vertical-align:bottom" width="68" alt="View the MathML source" title="View the MathML source" src="http://ars.els-cdn.com/content/image/1-
s2.0-S0378779612002891-si4.gif>"
where ni is the number of ensemble members, ɛ is the final standard deviation of error, which is defined as the difference between the input signal and the corresponding IMFs [22].
In practice, the number of ensemble members is often set to 100 and the standard deviation of white noise series is set to 0.1 or 0.2.
2.3 .Comparison of EMD and EEMD decomposition for feature extraction and selection
The extracted IMFs by EMD and EEMD are shown in Fig. 3a and b. The low-level IMF contains high-frequency components; while the high-level IMF contains low-frequency components.
<img class="figure large" border="0" alt="Full-size image (104 K)"
src="http://ars.els-cdn.com/content/image/1-s2.0-S0378779612002891-gr3.jpg" data-thumbsrc="http://ars.els-cdn.com/content/image/1-s2.0-S0378779612002891-gr3.sml" data-
fullsrc="http://ars.els-cdn.com/content/image/1-s2.0-S0378779612002891-gr3.jpg>"
Fig. 3. (a) Feature vectors (IMFs) for a voltage sag signal with EMD. (b) Feature vectors (IMFs) for a voltage sag signal with EEMD.
View thumbnail images
Figure options
View in workspace
Download full-size image
Download as PowerPoint slide
In Fig. 3a, the IMFs with EMD algorithm have side effects that reduce the accuracy of the classification algorithms. However, the EEMD algorithm does not present side effects and is,
therefore, used in this study. Another criterion is the Pearson product-moment correlation coefficient (R). It is a measure of the correlation (linear dependence) between two variables (X and
Y), giving a value between +1 and −1 inclusive.
An equivalent expression gives the correlation coefficient as the mean of the products of the standard scores. Based on a sample of paired data (Xi, Yi), the sample Pearson correlation coefficient
is defined in Eq. (5).
(5)
<img height="46" border="0" style="vertical-align:bottom" width="265" alt="View the MathML source" title="View the MathML source"
src="http://ars.els-cdn.com/content/image/1-s2.0-S0378779612002891-si5.gif>"
where <img height="18" border="0" style="vertical-align:bottom" width="87" alt="View the MathML source" title="View the MathML source"
src="http://ars.els-cdn.com/content/image/1-s2.0-S0378779612002891-si6.gif">, <img height="14" border="0" style="vertical-align:bottom" width="13" alt="View the MathML source"
title="View the MathML source" src="http://ars.els-cdn.com/content/image/1-s2.0-S0378779612002891-si7.gif"> and σX are the standard score, sample mean, and sample standard
deviation of X. Table 1 shows the R values for EMD and EEMD.
Table 1. Relation with first IMFs and current power signal.
Cross-CorrelationEMDEEMD
IMF–healthy state0.976
21.000
0
IMF–voltage sag0.881
90.908
1
IMF–voltage swell0.930
00.968
5
IMF–voltage flicker0.970
00.995
0
Cross-CorrelationEMDEEMD
IMF–harmonics (3, 5, 7)0.785
50.786
1
IMF–transient0.568
50.610
4
IMF–dc component0.924
40.935
7
IMF–EMI0.533
40.642
9
IMF–outage0.931
30.985
6
Full-size table
Table options
View in workspace
Download as CSV
Table 1 also proves that EEMD has better resolution for the analyzed PQ events.
2.4 .Hilbert-Huang Transformation Method (HHT)
HHT represents the original signal X(t) into the time frequency domain by combining the empirical mode decomposition with the Hilbert transform. The advantage of using HHT for feature extraction
procedure can be summarized in Table 2.
Table 2. Comparison of frequently used mathematical methods for feature extraction [20] and [21].
FourierSTFTWaveletHHT
BasisNon-adaptiveNon-adaptiveNon-adaptiveAdaptive
FrequencyConvolution:
GlobalConvolution:
RegionalConvolution:
RegionalDifferentiation:
Local
PresentationEnergy-
frequencyEnergy-time-
frequencyEnergy-time-
frequencyEnergy-time-
frequency
Non-linearNoNoNoYes
Non-stationaryNoNoYesYes
Feature Extraction
NoDiscrete: No
Continuous: YesDiscrete: No
Continuous: YesYes
FourierSTFTWaveletHHT
Theoretical base
Theory complete
Theory completeTheory completeempirical
Full-size table
Table options
View in workspace
Download as CSV
EMD can decompose signal into a finite number of n IMFs (Cj, j = 1,2,3...,n), which extract the energy associated with various intrinsic time scales and residual rn; this step is called the sifting process
described as in Eq. (6). IMFs, as a class of functions, satisfy two definitions. In the whole data set, the number of extrema and the number of zero-crossings must be either equal or differ at most by one.
At any point, the mean value of the envelope defined by the local maxima and minima is zero.
The Hilbert transform is then applied to each IMF component Cj:
(6)
<img height="40" border="0" style="vertical-align:bottom" width="171" alt="View the MathML source" title="View the MathML source"
src="http://ars.els-cdn.com/content/image/1-s2.0-S0378779612002891-si8.gif>"
where cj(t) and vj(t) are real part and imaginary part of an analytic signal zj(t):
(7)
<img height="46" border="0" style="vertical-align:bottom" width="168" alt="View the MathML source" title="View the MathML source"
src="http://ars.els-cdn.com/content/image/1-s2.0-S0378779612002891-si9.gif>"
With amplitude, (IA), and phase defined by the expressions:
(8)
<img height="76" border="0" style="vertical-align:bottom" width="167" alt="View the MathML source" title="View the MathML source"
src="http://ars.els-cdn.com/content/image/1-s2.0-S0378779612002891-si10.gif>"
Therefore, the instantaneous frequency (IF), was given by:
(9)
<img height="35" border="0" style="vertical-align:bottom" width="94" alt="View the MathML source" title="View the MathML source" src="http://ars.els-cdn.com/content/image/1-
s2.0-S0378779612002891-si11.gif>"
Thus, the original data can be expressed in the following form:
(10)
<img height="48" border="0" style="vertical-align:bottom" width="243" alt="View the MathML source" title="View the MathML source" src="http://ars.els-
cdn.com/content/image/1-s2.0-S0378779612002891-si12.gif>"
Fig. 4 and Fig. 5 show IA and IF curves for healthy and voltage sag signals. Fig. 6 shows the IA values for all analyzed PQ events.
<img class="figure large" border="0" alt="Full-size image (36 K)" src="http://ars.els-cdn.com/content/image/1-s2.0-
S0378779612002891-gr4.jpg" data-thumbsrc="http://ars.els-cdn.com/content/image/1-s2.0-S0378779612002891-gr4.sml" data-fullsrc="http://ars.els-cdn.com/content/image/1-s2.0-
S0378779612002891-gr4.jpg>"
Fig. 4. Instantaneous amplitude (IA) for healthy and voltage sag signal.
View thumbnail images
Figure options
View in workspace
Download full-size image
Download as PowerPoint slide
<img class="figure large" border="0" alt="Full-size image (36 K)" src="http://ars.els-cdn.com/content/image/1-s2.0-
S0378779612002891-gr5.jpg" data-thumbsrc="http://ars.els-cdn.com/content/image/1-s2.0-S0378779612002891-gr5.sml" data-fullsrc="http://ars.els-cdn.com/content/image/1-s2.0-
S0378779612002891-gr5.jpg>"
Fig. 5. Instantaneous frequency (IF) for healthy and voltage sag signal.
View thumbnail images
Figure options
View in workspace
Download full-size image
Download as PowerPoint slide
<img class="figure large" border="0" alt="Full-size image (34 K)" src="http://ars.els-cdn.com/content/image/1-s2.0-
S0378779612002891-gr6.jpg" data-thumbsrc="http://ars.els-cdn.com/content/image/1-s2.0-S0378779612002891-gr6.sml" data-fullsrc="http://ars.els-cdn.com/content/image/1-s2.0-
S0378779612002891-gr6.jpg>"
Fig. 6. IA corresponding to first IMF for all disturbances.
View thumbnail images
Figure options
View in workspace
Download full-size image
Download as PowerPoint slide
3 .Singular value decomposition (SVD)
SVD is used for calculating singular values and increases the accuracy of the classification algorithm. Singular values give more information for identification of the system. SVD is a part of hierarchical
classification on multiple feature groups in this study.
For any real m × n matrix A, there exist orthogonal matrices
(11)
<img height="45" border="0" style="vertical-align:bottom" width="200" alt="View the MathML source" title="View the MathML source" src="http://ars.els-
cdn.com/content/image/1-s2.0-S0378779612002891-si13.gif>"
such that:
(12)
A=UΣVT
where Σ=diag(σ1,σ2,...,σmin(m,n))) Rm×n and σ1≥σ2≥...≥σmin(m,n)≥0∈.
The σi is the ith singular value of A in non-increasing order and the vectors ui and vi are the ith left singular vector and the ith right singular vector of A for i = min(m,n), respectively [22]. The singular
values of matrix A are unique, while the singular vectors corresponding to distinct singular values are uniquely determined up to the sign.
Fig. 7 shows singular values for analyzed power signals. Fig. 7 only shows 3 sets of data over 90 training data sets. In Fig. 7, SV1 represents singular values for the healthy signal, SV2 for voltage sag,
SV3 for voltage swell, SV4 for flicker, SV5 for harmonics, SV6 for transients, SV7 for DC component, SV8 for EMI and finally SV9 for the voltage interruption signal.
<img class="figure large" border="0" alt="Full-size image (46 K)"
src="http://ars.els-cdn.com/content/image/1-s2.0-S0378779612002891-gr7.jpg" data-thumbsrc="http://ars.els-cdn.com/content/image/1-s2.0-S0378779612002891-gr7.sml" data-
fullsrc="http://ars.els-cdn.com/content/image/1-s2.0-S0378779612002891-gr7.jpg>"
Fig. 7. Singular values for training data set.
View thumbnail images
Figure options
View in workspace
Download full-size image
Download as PowerPoint slide
4 .Support vector machines (SVMs)
SVMs are a class of learning machines that find optimal hyper-planes among different classes of input data or training data in a high dimensional feature space F, and new test data can be classified
using the individual hyper-planes. The optimal hyper-plane, found during a training phase, makes the smallest number of training errors [23]. Fig. 8 illustrates an optimal hyper-plane for two classes
of training data.
<img class="figure large" border="0" alt="Full-size image (28 K)" src="http://ars.els-cdn.com/content/image/1-s2.0-
S0378779612002891-gr8.jpg" data-thumbsrc="http://ars.els-cdn.com/content/image/1-s2.0-S0378779612002891-gr8.sml" data-fullsrc="http://ars.els-cdn.com/content/image/1-s2.0-
S0378779612002891-gr8.jpg>"
Fig. 8. Optimal hyperplane.
View thumbnail images
Figure options
View in workspace
Download full-size image
Download as PowerPoint slide
Let {xi,yi}, i = 1,2,…,N be N training data vectors xi with class label yi. Given an input vector x, an SVM constructs a classifier of the form:
(13)
<img height="49" border="0" style="vertical-align:bottom" width="233" alt="View the MathML source" title="View the MathML source" src="http://ars.els-
cdn.com/content/image/1-s2.0-S0378779612002891-si17.gif>"
where {ai} are non-negative Lagrange multipliers, each of which corresponds to a training data, b is a bias constant, and K(.,.) is a kernel satisfying the conditions of Mercer's theorem [24] and [25].
Frequently used kernel functions are the polynomial kernel K(xi,xj)=d(xi.xj+1) and Gaussian Radial Basis Function (RBF), K(xi,xj)=e−|xi−xj|2/2σ2.
The problem of finding the optimal hyperplane is specified by the following quadratic programming problem:
Minimizing:
(14)
<img height="51" border="0" style="vertical-align:bottom" width="297" alt="View the MathML source" title="View the MathML source"
src="http://ars.els-cdn.com/content/image/1-s2.0-S0378779612002891-si20.gif>"
subject to:
(15)
<img height="48" border="0" style="vertical-align:bottom" width="79" alt="View the MathML source" title="View the MathML source" src="http://ars.els-cdn.com/content/image/1-
s2.0-S0378779612002891-si21.gif>"
0≤ai≤C, i=1,2,...,N
The above quadratic programming problem can be solved with traditional optimization techniques. The vectors for which ai > 0 after optimization are called support vectors.
4.1 .One-against-all method
For a k-class problem, the one-against-all method constructs k SVM models. The ith SVM is trained with all of the training examples in the ith class with positive labels and the others with negative
labels. The final output of the one-against-all method is the class that corresponds to the SVM with the highest output value. Given a set of training examples D{(x1,y1),(x2,y2),...,(xN,yN)}, where the ith sample xi∈Rn and yj∈{1,...,k} is the class of xj the ith SVM solves the following optimization problem
[26], [27] and [28] (where c > 0 is a regularization parameter for the tradeoff between model complexity and training error, and ξi measures the (absolute) difference between f(x)=T(wi)ϕ(x)+bi
and yi.):
(16)
<img height="48" border="0" style="vertical-align:bottom" width="158" alt="View the MathML source" title="View the MathML source"
src="http://ars.els-cdn.com/content/image/1-s2.0-S0378779612002891-si27.gif>"
subject to
(17)
<img height="87" border="0" style="vertical-align:bottom" width="249" alt="View the MathML source" title="View the MathML source" src="http://ars.els-
cdn.com/content/image/1-s2.0-S0378779612002891-si28.gif>"
The decision function of the ith SVM is
(18)
<img height="19" border="0" style="vertical-align:bottom" width="148" alt="View the MathML source" title="View the MathML source"
src="http://ars.els-cdn.com/content/image/1-s2.0-S0378779612002891-si29.gif>"
A point x is in the class that corresponds to the largest value of the decision functions: the class of
(19)
<img height="19" border="0" style="vertical-align:bottom" width="223" alt="View the MathML source" title="View the MathML source" src="http://ars.els-
cdn.com/content/image/1-s2.0-S0378779612002891-si30.gif>"
A comparative study is carried out in the next section that compares the performances of the OAO (one-against-one) and OAA (one-against-all) decomposition methods in training the proposed
multiclass SVMs. The results of the application of the proposed algorithm to the data set are presented in Table 3.
Table 3. Comparisons of the performances of the one-against-all (OAA) and the one-against-one (OAO) SVM classifiers.
Total number of training/testing segments
Classification (SVM) One Against All (OAA)–One Against One (OAO)
90/25HSVSGVSWVFVHVTRVDCEMIVI
HS3–200000000
VSG12–20000000
VSW002–2100000
VF0003–200000
VH00003–30000
VTR000004–3010
VDC0000003–300
EMI00000103–20
VI000000002–2
HS: healthy state, VSG: voltage sag, VSW: voltage swell, VF: voltage flicker, VH: voltage harmonics, VTR: voltage transients, VDC: DC voltage, EMI: electro magnetic interference.
Full-size table
Table options
View in workspace
Download as CSV
Diagonal elements represent the correctly classified PQ disturbances while non-diagonal elements represent incorrect classification results (Table 3).
5 .Experimental results and discussions
To evaluate the performance of the proposed power quality classification algorithm, a total of 115 PQ disturbance data were used. The PQ signals in each class were divided into 90 training sets and
25 test sets. The polynomial kernel is used as the kernel function and the parameter d is chosen as 2 in the SVM classifier. The exact number of the training and testing segments for each PQ event is
shown in Table 2. Testing data were obtained from TUBITAK. This is an ongoing project and is devoted to taking the nationwide power quality figures of the Turkish Electricity Transmission
System through field measurements carried out by application-specific mobile monitoring equipment. In this project, PQ measurements are being carried out according to IEC 61000–4-30,
except for a continuous seven-day sampling rate in the selected locations. In the transmission system, most of the bus bars have more than one incoming and outgoing feeder. Therefore, the
number of measurement points can be taken as the number of feeders for currents and as the number of bus bars for voltages. The classification of these points as heavy industry, urban +
industrial, and urban sites are given in Table 4, with respect to voltage level. It is noted that there are more than 95 incoming and outgoing feeders at 33 measured transformer substations [29], [30] and
[31]. The real-time data are acquired in 3 s, which corresponds to 150 periods.
Table 4. Number of measurement point [29], [30] and [31].
Number of measurement points
Voltage levelLoad type
Heavy IndustryIndustry + urbanUrban onlyTotal
NBNFNBNFNBNFNB
NF
33 kV816172814163960
154 kV351013992227
Number of measurement points
Voltage levelLoad type
Heavy IndustryIndustry + urbanUrban onlyTotal
NBNFNBNFNBNFNB
NF
380 kV11470058
Total1222314823256695
NB: number of buses, NF: number of feeders.
Full-size table
Table options
View in workspace
Download as CSV
The most accurate multi classifiers (OAA) for each PQ dataset are indicated by red font. As seen from Table 2, the OAA method (Error = 0%, Precision = 100%) is superior to the OAO method (Error = 16%, Precision = 84%). Therefore, the OAA decomposition method is chosen for the multiclass SVM in this
study. According to the results of the test data, the effectiveness of the proposed SVM algorithm is made suitable by increasing the RBF kernel parameter's standard deviation (sigma 0.1 to 1). The
polynomial kernel is worse than the RBF kernel due to the CPU time. It has been observed that additional properties of the feature vectors such as singular values increase the performance of
classification algorithm. The performance of SVM algorithm with OAA and OAO is calculated as 100% and 84%, respectively. Nevertheless, the performance of SVM algorithm decreases dramatically
without using singular values of IF and IA values and is calculated as 80% with OAA and 56% with OAO, respectively. Table 5 gives a comparison of the different classifiers on the training and testing
data.
Table 5. Comparison of performances of the different classifiers.
ClassifierErrorAccuracyTime (s)
SVM-Linear0.400
00.60002.136
SVM-Poly d = 20.0001.00012.711
ClassifierErrorAccuracyTime (s)
SVM-RBF sigma = 0.10.280
00.72008.208
SVM-RBF sigma = 10.000
01.00010.686
Gauss Mixture Model0.255
50.74441.493
AdaBoost + Naive Bayes0.040
00.96003.529
Bagging + Naive Bayes0.160
00.84002.577
Full-size table
Table options
View in workspace
Download as CSV
6 .Conclusion
In this paper, a different PQ monitoring approach based on EEMD, HHT, and SVM was proposed. EEMD was used for signal conditioning technique and two distinctive features, IA and IF, were
obtained by HHT. Then a series of statistical calculations was performed to form feature matrix. The algorithm is found capable of classifying many PQ disturbances with a high accuracy.
SVM with radial basis (sigma = 1) kernel function was used for the decision-making unit and presented high accuracy among the other classifiers. The proposed algorithm was applied to two
sets of data, one for generated data on a computer and the other for the data obtained from TUBITAK. Practical applications showed that the proposed hybrid algorithm can be used in power
quality monitoring software.
The analysis and results presented in this work clearly show the potential capability of the suggested hybrid algorithm in classifying the distorted PQ waveforms.
Acknowledgement
The authors would like to thank TUBITAK for supplying real-time data.
References
[1]
M.D.L. Angrisani, P. Daponte, A. Testa
A measurement method based on the wavelet transform for power quality analysis
IEEE Transactions on Power Delivery, 13 (4) (1988), pp. 990–998
[SD-008]
[2]
D. Granados-Lieberman, R.J. Romero-Troncoso, R.A. Osornio-Rios, A. Garcia-Perez, E. Cabal-Yepez
Techniques and methodologies for power quality analysis and disturbances classification in power systems: a review
IET Generation, Transmission & Distribution, 5 (4) (2011), pp. 519–529
[SD-008]
[3]
C. Muscas
Power quality monitoring in modern electric distribution systems
IEEE Instrumentation and Measurement Magazine, 13 (2010), pp. 19–27
[SD-008]
[4]
P.J. Loughlin, J.W. Pitton, E. Les
Atlas, Proper time-frequency energy distributions and the Heisenberg uncertainty principle
Proceedings of The IEEE-SP International Symposium, Victoria, BC, Canada (1992), pp. 151–154
[SD-008]
[5]
I. Daubechies
Ten Lectures on Wavelets
SIAM, Philadelphia, PA (1992)
[SD-008]
[6]
S. Mallat
A theory for multiresolution signal decomposition: the wavelet representation
IEEE Transactions on Pattern Analysis, 11 (1989), pp. 674–693
[SD-008]
[7]
R.R. Coifman, M.V. Wickerhauser
Entropy-based algorithms for best basis selection
IEEE Transactions on Information Theory, 38 (1992), pp. 713–718
[SD-008]
[8]
P. Kailasapathi, D. Sivakumar
Methods to analyze power quality disturbances
European Journal of Scientific Research, 47 (1) (2010), pp. 6–16
[SD-008]
[9]
P.K. Dash, R.K. Jena, M.M.A. Salama
Power quality monitoring using an integrated Fourier linear combiner and fuzzy expert system
International Journal of Electrical Power, 21 (1999), pp. 497–506
[SD-008]
[10]
S. Suja, Jovitha Jerome
Pattern recognition of power signal disturbances using S Transform and TT Transform
International Journal of Electrical Power, 32 (2010), pp. 37–53
[SD-008]
[11]
P.K. Dash, S.K. Panda, A.C. Liew, B. Mishra, R.K. Jena
A new approach to monitoring electric power quality
Electric Power Systems Research, 46 (1998), pp. 11–20
[SD-008]
[12]
M. Oleskovicz, D.V. Coury, O.D. Felho, W.F. Usida, A. driano, A.F.M. Carneiro, Leandro R.S. Pires
Power quality analysis applying a hybrid methodology with wavelet transforms and neural Networks
Electric Power Systems Research, 31 (2009), pp. 206–212
[SD-008]
[13]
S.A. Soliman, M.E. El-Hawary
Measurement of power systems voltage and flicker levels for power quality analysis: a static LAV state estimation based algorithm
International Journal of Electrical Power, 22 (2000), pp. 447–450
[SD-008]
[14]
S.A. Soliman, A.H. Mantaway, M.E. El-Hawary
Simulated annealing optimization algorithm for power systems quality analysis
International Journal of Electrical Power, 26 (2004), pp. 31–36
[SD-008]
[15]
M. Aiello, A. Cataliotti, S. Nuccio
A Chirp-Z transform-based synchronizer for power system measurements
IEEE Transactions on Instrumentation and Measurement, 54 (2005), pp. 1025–1032
[SD-008]
[16]
G. Zheng, Shi Mei-Xiang, D. Liu, J. Yao, Miao Zhu-Mei
Power quality disturbance classification based on rule-based and wavelet-multi-resolution decomposition
Proc. 2002 Int. Conf. Mach. Learn. Cybernetics, 4 (2002), pp. 2137–2141
[SD-008]
[17]
M. Uyar, S. Yildirim, M. Tunay Gencoglu
An effective wavelet-based feature extraction method for classification of power quality disturbance signals
Electric Power Systems Research, 78 (2008), pp. 1747–1755
[SD-008]
[18]
S. Kaewarsa, K. Attakitmongcol, T. Kulworawanichpong
Recognition of power quality events by using multiwavelet based neural Networks
Electric Power Systems Research, 30 (2008), pp. 254–260
[SD-008]
[19]
T. Jayasree, D. Devaraj, R. Sukanesh
Power quality disturbance classification using Hilbert transform and RBF Networks
Neurocomputing, 73 (2010), pp. 1451–1456
[SD-008]
[20]
N.E. Huang, Z. Shen, S.R. Long, M.C. Wu, H.H. Shih, Q. Zheng, N.C. Yen, C.C. Tung, H.H. Liu
The empirical mode decomposition and the Hilbert spectrum for nonlinear and non-stationary time series analysis
Proceedings of the Royal Society A-Mathematical Physical and Engineering Sciences, 454 (1998), pp. 903–995
[SD-008]
[21]
Z. Wu, N.E. Huang
Ensemble empirical mode decomposition: a noise-assisted data analysis method
Advances in Adaptive Data Analysis, 1 (2009), pp. 1–41
[SD-008]
[22]
G.H. Golub, C.F.V. Loan
Matrix Computations
The Johns Hopkins University Press, Baltimore, MD (1996)
[SD-008]
[23]
C. Cortes, V. Vapnik
Support-vector networks
Machine Learning, 20 (1995), pp. 273–297
[SD-008]
[24]
V.N. Vapnik
Statistical Learning Theory
Wiley, New York (1998)
[SD-008]
[25]
Chih-Wei Hsu, Chih-Jen Lin
A comparison of methods for multiclass support vector machines
IEEE Transactions on Neural Networks, 13 (2) (2002), pp. 415–425
[SD-008]
[26]
J.C. Platt, N. Cristianini, J. Shawe-Taylor
Large margin DAGs for multi-class classification
Advances in Neural Information Processing Systems, 12 (2000), pp. 547–553
[SD-008]
[27]
C.W. Hsu, C.J. Lin
A comparison of methods for multi-class support vector machines
IEEE Transactions on Neural Networks, 13 (2002), pp. 415–425
[SD-008]
[28]
R. Rifkin, A. Klautau
In defense of one-vs-all classification
Journal of Machine Learning Research, 5 (2004), pp. 101–141
[SD-008]
[29]
E. Ozdemirci, Y. Akkaya et al.
Mobile monitoring system to take PQ snapshots of Turkish electricity transmission system
Instrumentation and Measurement Technology Conference Proceedings, British Columbia, Canada (2008), pp. 1–6
[SD-008]
[30]
T. Demirci, A. Kalaycioglu et al.
National power quality monitoring network and assessment center for the Turkish electricity transmission system: recent developments
Signal Processing, Communication and Applications Conference, Antalya, Turkey (2008), pp. 1–4
[SD-008]
[31]
http://www.guckalitesi.gen.tr/en/about/scope.php (last accessed 22.07.2012)
[SD-008]
Corresponding author. Tel.: +90 362 4377204.
Crown copyright © 2012 Published by Elsevier B.V. All rights reserved.
Bibliographic information
Citing and related articles
Related articles
No articles found.
Cited by in Scopus (0)
Related reference work articles
No articles found.
Applications and tools
Workspace
No content has been selected.
About ScienceDirect
About Elsevier
Contact and support
Information for advertisers
Terms and conditions
Privacy policy
Copyright © 2012 Elsevier B.V. All rights reserved. SciVerse® is a registered trademark of Elsevier Properties S.A., used under license. ScienceDirect® is a registered trademark of Elsevier B.V .
Top of Form
Bottom of Form
ScienceDirect article suggestions
Related articles
These articles have key terms similar to those in the article you downloaded.Learn more
Do not show again
A new classif ication for pow er quality events in distribution systems¶Okan Ozgonenel, Turgay Yalcin, Irfan Guney, Unal Kurt¶Published in:¶Electric Pow er Systems Research¶Volume 95, February 2013, Pages 192-199¶doi: 10.1016/j.epsr.2012.09.007