POWER QUALITY DISTURBANCE ANALYSIS USING S-TRANSFORM AND DATA MINING BASED CLASSIFIER

POWER QUALITY DISTURBANCE ANALYSIS USING S-TRANSFORM AND DATA MINING BASED CLASSIFIER Swarnabala Upadhyaya 1 and Ambarish Panda 2 1,2 Department of Electrical Engineering SUIIT,Sambalpur Odisha-768019, India Abstract- In the recent years, multiresolution based methodologies have been developed as suitable alternative for real-time signal analysis. This paper has implemented S-transform for analysis of different type of power quality (PQ) disturbance signals. Ten types of different PQ events of the voltage signal such as sag, swell, interruption, harmonic, sag with harmonics and swell with harmonics, spike, notch etc. are analyzed through the S-contours. Suitable features have been extracted from the S-matrix. The extracted features have been fed to data mining based Decision tree (DT) classifier in order to discriminate the PQ distur bances. These aforementioned methods are also implemented on real signals. Keywords- Power quality disturbance, S-transform, Decision tree, Classification accuracy I. INTRODUCTION The continuous growth in the application of microprocessor-based controls and power electronic devices and adjustable-speed motor drives increases emphasis on quality of power as these are more sensitive to power quality variations than the traditional equipments. So, the term power quality has become an prolific buzzwords in the power industry since t he late 1980s. Moreover, Power quality (PQ) is like a umbrella which covers various disturbances of voltage and current such as voltage sag, swell, harmonics and oscillatory transients cause mal-function of sophisticated equipment. In other words the power quality is a nonstop dynamic variation both on the time and space. The concern over quality of power has been increasing rapidly due to dependance of present life on continuous supply of electrical energy. Similarly, the continuous increasing of load demand both in public sectors as well as industries has made the PQ as an serious issue. The presence of disturbances in the loads creates the deviation of voltage and current from the ideal waveform that declines the performance and the lifespan of equipments and also generate instabilities in system and so on. Hence,the healthy power system operation requires continues supervision, proper monitoring and optimum control in term of power quality improvement. In order to identify the type of power signal waveforms, researchers have been proposed different methodology such as the Fourier transform (FT), the short-time Fourier transform (STFT), the wavelet transform (WT) [1], [2], [3], and [4]. The commonly used STFT is only suitable for steady state disturbances but not for dynamic disturbances like transient due to the fixed window [5], [6] and [7]. Heisenberg-Gabor inequality [ 8] limits the time-frequency resolution of the signals in STFT analysis. However, multi resolution analysis (MRA) based WT is extensively used for nonstationary signal characterization. Wavelet Transform employees small wavelet. Wavelet can be defined as a oscillatory function having zero mean (no d.c component)and decaying to zero. WT uses the basis function known as mother wavelet unlike the Fourier transform (FT). The automatic detection of PQ events with the discrete wavelet transform (DWT) is a common topic in past studies [9], But the application of DWT is restricted with the size of the signal. The DWT decomposition provides the time scale representation. The extension of the wavelet idea is based on a moving and scalable localizing Gaussian window known as the S-transform. The automatic and the fast characterization of the different power quality disturbance signals has been DOI:10.21884/IJMTER.2018.5102.LCJWW 46

become an emerging issue for the power system researchers. Some common automated classification models are based on the Artificial Neural Network (ANN) [11], [12], fuzzy and neuro-fuzzy systems [13], [14], [15]. But the ANN suffers from the more number of training cycles which makes a huge computational burden. However, the main disadvantage of the traditional ANN based classifier is the requirement of retraining when a new phenom enon is added. Similarly, the Hidden Markov Model (HMMs) classifier is fails to classify the slow phenomena like inter ruption, sag etc properly [16] and [17]. The paper is organised as follows. Section II describes the brief theory of the S-transform for carrying out the process of localisation. The feature extraction is given in Section III. Section IV provides brief idea about the decision tree. The PQ disturbance model is given in Section V. The effectiveness of S-transform and decision tree has been presented for discriminate different PQ disturbances in Section V I and Section V II. Finally the Section V III, provides concluding remark. II. S-TRANSFORM The S-transform is the derived form of the continuous wavelet transform with a phase correction factor. In other words, the S-transform is the combination of the WT and short time Fourier transform that provides the time-frequency spectral localization of the signals. The frequency dependent variable window provides multiresolution analysis(mra) while retaining the absolute phase of each frequency. So, the S-transform provides proper detection and identification of time series si gnal of PQ. A. S-transform approach in power quality The multiresolution analysis of the S-transform makes it as a suitable tool for time series analysis in power system environment [18]. Mathematically, the S-transform of a continuous time signal s(t) is presented in equation(1) as given by where f is the frequency, t is the time and the ζ is the control parameter that controls the Gaussian window position on the t-axis. The factor α controls the time and the frequency resolution. As a result the frequency resolution increases, when the parameter α value is above 1.Similarly, if α decreases below 1, the time resolution improves [19]. In this work, the α is taken as.5 for analysis of all these aforementioned signals. A power signal (space between A and power should reduce)power signal h(t) in discrete form is expressed as h(kt ), for k = 1, 2,..., N 1 and the sampling time interval T. Mathematically, the discrete version of Fourier transform of the h(kt ) is given in [20] and expressed by equation (2) @IJMTER-2018, All rights Reserved 47

The (4) provides zero frequency voice. The output of the S-transform is an N M matrix is known as S-matrix. The row of the S-transform represents the frequency and the column represent the time. Moreover, each element of the matrix is a complex value. The averaging of the amplitude of the S-matrix over time results in Fourier spectrum [21]. In this work the α value is.5 for localization of both the stationary and the non-stationary signals. III. FEATURE EXTRACTION From the S-matrix matrix tree features are extracted. A feature such as energy, entropy and standard deviation has been extracted by the given equations [22], [23], [24]. where N is the number of samples in row of S-matrix. These extracted features are further fed to classifier reduce the computational burden of the raw data. The classification alg orithms are discussed in the next section. Figure. 1: Structure of DT IV. DECISION TREE (DT) A decision tree is a flowchart-like tree structure which can b e designed from top to down, bottom to up and other special approaches. However, top to down approach is commonly accepted and generally are drawn from left to right. A tree consists of three parts the root node, the internal node and the leaf node. A node maps a certain characteristic and the branches carrya range of values [25]. The basic block diagram of a DT is presented in Fig. 1 @IJMTER-2018, All rights Reserved 48

Root node: In this node, the operation of DT starts with the entire data samples. Internal node: The next step is the division of the records according to their features. The assigned node is called the internal node. Leaf node: Similarly, the next step is assignment of a class label to the nodes. The class label assigned nodes are called leaf node. The most widely used top-down algorithm of DT has been presented. V. POWER QUALITY DISTURBANCE MODEL The PQ analysis comprises of both the stationary as well as non-stationary signals such as the voltage swell, interruption, spike, sag, notch and so on. In this paper, ten types of different disturbances along with the pure sine wave are considered for analysis. These PQ disturbances are analysed with ten cycles of a waveform of 50 Hz fundamental frequency. The sampling frequency is 3.2 KHz. The signals are simulated according to the model [26] and [27]. VI. RESULTS AND DISCUSSION ON DETECTION The S-transform provides high frequency resolution at low frequency and high time resolution at high frequency. The MRA based S-transform is employed on the PQ signals in order to get localize the disturbance. The equation (1) to equation(4) are implemented to localize ten types of PQ disturbances along with the pure sinusoidal signals. A. Pure Sinusoidal Wave A pure sinusoidal wave of voltage signal is considered in Fig.2. With S-transform the pure signal is analysed are shown in Fig.2. Similarly, the pure sine wave voltage signal is considered for analysis with ST in Fig. 2. The vertical axis presents the frequency in khz and the horizontal axis presents the the time (in second) in terms of samples. As it is distortion free signal so there is deviation in S-contour. Table 1: Power quality Disturbance Models @IJMTER-2018, All rights Reserved 49

Figure. 2: Localization of pure sine wave using S-transform B. Pure Sinusoidal Wave with Sag The voltage signal with Sag is considered in Fig. 3. The voltage dip is properly identified from the time frequency plot of the S-transform contours. The sag is clearly localized by the contours. The contours has provided reduction in magnitude during the disturbance similar to the sag in the voltage signal. C. Pure Sinusoidal Wave With Swell Similarly, the swell in the sinusoidal voltage signal is localized by the increased magnitude of the contours. The patters produces a swell in the magnitude during the distortion and is given in Fig. 4. D. Pure Sinusoidal Wave with Interruption The huge reduction in the magnitude of the contours similar to the interruption in the voltage is shown in Fig. 5. From the Fig. 3 and Fig. 5 it can be observed that the magnitude of contours of interruption is different from the sag. E. Pure Sinusoidal Wave With Notch The signal with notch in each cycle is localized by the highly increased magnitude of contour and is shown in Fig. 6. Figure. 3: Localization of sag in pure sine wave using S-transform @IJMTER-2018, All rights Reserved 50

Figure. 4: Localization of swell in pure sine wave using S-transform Figure. 5: Localization of interruption in pure sine wave using S-transform F. Pure Sinusoidal Wave With Oscillatory Transients The S-transform is implemented on the oscillatory transient signal and Fig. 7 shows the distortion getting localized in the S-transform contours. G. Pure Sinusoidal Wave With Flicker The flicker signal is considered for analysis in Fig. 8. H. Pure Sinusoidal Wave With Spike Similar to the notch signal, the spike at each cycle of pure sine wave is also considered for analysis. Similar phenomena is applicable for the spike given in Fig. 9. I. Pure Sinusoidal Wave With Harmonics The sine wave with harmonic is analyzed in ST implementation presented in Fig. 10. The harmonic is identified by the contours of the ST. J. Pure Sinusoidal Wave With Sag and Harmonics The reduced magnitude contours corresponds to the sag in voltage shown in Fig. 1 Figure. 6: Localization of notch in pure sine wave using S-transform @IJMTER-2018, All rights Reserved 51

Figure. 7: Localization of oscillatory transient in pure sine wave using S-transform Figure. 8: Localization of flicker in pure sine wave using S-tran sform Figure. 9: Localization of spike in pure sine wave using S-transform Figure. 10: Localization of harmonic in pure sine wave using S-transform @IJMTER-2018, All rights Reserved 52

Figure. 11: Localization of harmonic in pure sine wave using S-transform Figure. 12: Localization of harmonic in pure sine wave using S-transform K. Pure Sinusoidal Wave With Swell and Harmonics The swell with harmonic in sine wave is properly identified by the increased magnitude of the contours. This is presented in Fig. 12. Similarly the reduced magnitude contours corresponds to the sag in voltage shown in Fig. 11 The aforementioned PQ signals have simulated with a core-i5,2.40 GHz under MATLAB environment. The PQ disturbance are clearly localised in the waveform. All distortions are not properly identified. So the classification is carried out i n order to discriminate different types of PQ events. VII. CLASSIFICATION RESULT A. Classification of synthesized PQD signals The Table II shows the computed values of the % CA of the classifier and misclassification rate (% MC) for each of the disturbance classes. For each class of disturbances, 125 signals are synthesized. In order to test the classifier, the classifier is Table 2: MC (%) and CA (%) of synthesized signals MC (%) or CA PQD events DT (%) % MC Sag 0 % MC Swell 0 % MC Interruption 0 @IJMTER-2018, All rights Reserved 53

% MC Oscillatory Transient 0 % MC Flicker 0.23 % MC Harmonics 0 % MC Sag+ Harmonics 0.08 % MC Swell+ Harmonics 0.46 % MC Notch 0 % MC Spike 0.78 % CA Total 98.93 Table 3: MC (%) and CA (%) of real signal Figure. 13: Experimental set up of signal generation @IJMTER-2018, All rights Reserved 54

(a) (b) Figure. 14: (a) Swell (b) Sag,swell with interruption Implemented in real environment B. Classification of Real PQD signals PQD events such as sag, swell, sag with swell and interruption are captured from a transmission panel of 220KV. The experimental set of the PQD signal generation is shown in Fig.13. Signal such as swell, sag plus interruption and sag, swell with interruption are shown in Fig.14. The signals are then extracted from oscilloscope and feed to MATLAB as input for feature extraction and subsequent classification. In order to validate the aforementioned methods, the generated four signals are fed for classification. For each class of disturbances, 55 signals are captured. From the Table III, it is observed that the aforementioned methods perform well with the real and synthesized signals. VIII. CONCLUSION The S-transform is a suitable tool for the analysis PQ disturbances signals. All the aforementioned ten types of PQ disturbances along with the sinusoidal voltage signal wave form are properly analysed with ST. The distortions are properly localized in the S-contours. Though harmonic is stationary, it has been properly realised from the S-contours. The data mining based classifier is also satisfactorily working in real envi ronment. Moreover, MRA based ST is a suitable tool for analysis of both stationary and non-stationary signals. REFERENCES [1] L. Coppola, Q. Liu, S. Buso, D. Boroyevich, and A. Bell, W avelet transform as an alternative to the short-time fourier transform for the study of conducted noise in power electronics, Industrial Electronics, IEEE Transactions on, vol. 55, no. 2, pp. 880 887, 2008. [2] Z.-L. Gaing, Wavelet-based neural network for power di sturbance recognition and classification, Power Delivery, IEEE Transactions on, vol. 19, no. 4, pp. 1560 1568, 2004. [3] J. G. Decanini, M. S. Tonelli-Neto, F. C. Malange, and C. R. Minussi, Detection and classification of voltage disturb ances using a fuzzy-artmap-wavelet network, Electric Power Systems Research, vol. 81, no. 12, pp. 2057 2065, 2011. [4] J. Barros, M. de Apraiz, and R. I. Diego, A virtual measur ement instrument for electrical power quality analysis using wavelets, Measurement, vol. 42, no. 2, pp. 298 307, 2009. [5] A. A. Abdelsalam, A. A. Eldesouky, and A. A. Sallam, Char acterization of power quality disturbances using hybrid technique of linear kalman filter and fuzzy-expert system, Electric power systems Research, vol. 83, no. 1, pp. 41 50, 2012. [6] A. Gaouda, M. Salama, M. Sultan, A. Chikhani, et al., Power quality detection and classification using wavelet - multiresolution signal decomposition, IEEE Transactions on Power Delivery, vol. 14, no. 4, pp. 1469 1476, 1999. [7] L. Angrisani, P. Daponte, M. Apuzzo, and A. Testa, A meas urement method based on the wavelet transform for power quality analysis, Power Delivery, IEEE Transactions on, vol. 13, no. 4, pp. 990 998, 1998. [8] B. Biswal and S. Mishra, Power signal disturbance ident ification and classification using a modified frequency slice wavelet transform, Generation, Transmission & Distribution, IET, vol. 8, no. 2, pp. 353 362, 2014. [9] A. G. Hafez, E. Ghamry, H. Yayama, and K. Yumoto, A wavele t spectral analysis technique for automatic detection of geomagnetic sudden commencements, Geoscience and Remote Sensing, IEEE Transactions on, vol. 50, no. 11, pp. 4503 4512, 2012. @IJMTER-2018, All rights Reserved 55

[10] D. B. Percival and A. T. Walden, Wavelet methods for tim e series analysis (cambridge series in statistical and probabilistic mathematics), 2000. [11] A. S. Yilmaz, A. Subasi, M. Bayrak, V. M. Karsli, and E. Ercelebi, Application of lifting based wavelet transforms t o characterize power quality events, Energy conversion and management, vol. 48, no. 1, pp. 112 123, 2007. [12] A. K. Ghosh and D. L. Lubkeman, The classification of pow er system disturbance waveforms using a neural network approach, Power Delivery, IEEE Transactions on, vol. 10, no. 1, pp. 109 115, 1995. [13] S. Hasheminejad, S. Esmaeili, and S. Jazebi, Power qua lity disturbance classification using s-transform and hidd en markov model, Electric Power Components and Systems, vol. 40, no. 10, pp. 1160 1182, 2012. [14] M. B. I. Reaz, F. Choong, M. S. Sulaiman, F. Mohd-Yasin, and M. Kamada, Expert system for power quality disturbance c lassifier, Power Delivery, IEEE Transactions on, vol. 22, no. 3, pp. 1979 1988, 2007. [15] S. Mishra, C. Bhende, and B. Panigrahi, Detection and c lassification of power quality disturbances using s-transf orm and probabilistic neural network, Power Delivery, IEEE Transactions on, vol. 23, no. 1, pp. 280 287, 2008. [16] M. Masoum, S. Jamali, and N. Ghaffarzadeh, Detection a nd classification of power quality disturbances using discr ete wavelet transform and wavelet networks, IET Science, Measurement & Technology, vol. 4, no. 4, pp. 193 205, 2010. [17] T. Abdel-Galil, E. El-Saadany, A. Youssef, and M. Salama, Disturbance classification using hidden markov models a nd vector quantization, Power Delivery, IEEE Transactions on, vol. 20, no. 3, pp. 2129 2135, 2005. [18] R. G. Stockwell, L. Mansinha, and R. Lowe, Localizatio n of the complex spectrum: the s transform, Signal Processing, IEEE Transactions on, vol. 44, no. 4, pp. 998 1001, 1996. [19] I. W. Lee and P. K. Dash, S-transform-based intelligen t system for classification of power quality disturbance sig nals, Industrial Electronics, IEEE Transactions on, vol. 50, no. 4, pp. 800 805, 2003. [20] P. Dash, B. Panigrahi, and G. Panda, Power quality anal ysis using s-transform, Power Delivery, IEEE Transactions on, vol. 18, no. 2, pp. 406 411, 2003. [21] S. Mishra, C. Bhende, and B. Panigrahi, Detection and c lassification of power quality disturbances using s-transf orm and probabilistic neural network, Power Delivery, IEEE Transactions on, vol. 23, no. 1, pp. 280 287, 2008. [22] C.-Y. Lee and Y.-X. Shen, Optimal feature selection fo r power-quality disturbances classification, Power Delivery, IEEE Transactions on, vol. 26, no. 4, pp. 2342 2351, 2011. [23] M. Biswal and P. K. Dash, Measurement and classificatio n of simultaneous power signal patterns with an s- transform variant and fuzzy decision tree, Industrial Informatics, IEEE Transactions on, vol. 9, no. 4, pp. 1819 1827, 2013. [24] B. Panigrahi and V. R. Pandi, Optimal feature selectio n for classification of power quality disturbances using wav elet packet-based fuzzy k-nearest neighbour algorithm, IET generation, transmission & distribution, vol. 3, no. 3, pp. 296 306, 2009. [25] G. Williams, Data mining with Rattle and R: the art of excavating data for knowledge discovery. Springer Science & Business Media, 2011. [26] S. Upadhyaya and S. Mohanty, Power quality disturbanc e detection using wavelet based signal processing, in India Conference (INDICON), 2013 Annual IEEE, pp. 1 6, IEEE, 2013. [27] S. Upadhyaya, S. Mohanty, and C. Bhende, Hybrid method s for fast detection and characterization of power quality disturbances, Journal of Control, Automation and Electrical Systems, vol. 26, no. 5, pp. 556 566, 2015. @IJMTER-2018, All rights Reserved 56