Paper Title (use style: paper title) TRANSACTIONS ON ENVIRONMENT AND ELECTRICAL ENGINEERING ISSN 2450-5730 Vol 1, No 3 (2016) © Turgay Yalcin, Muammer Ozdemir Pattern Recognition Methods for Detecting Voltage Sag Disturbances and Electromagnetic Interference in Smart Grids Turgay Yalcin, Muammer Ozdemir Abstract— Identification of system disturbances, detection of them guarantees smart grids power quality (PQ) system reliability and provides long lasting life of the power system. The key goal of this study is to find the best accuracy of identification algorithm for non-stationary, non-linear power quality disturbances such as voltage sag, electromagnetic interference in smart grids. PQube, power quality and energy monitor, was used to acquire these distortions. Ensemble Empirical Mode Decomposition is used for electromagnetic interference reduction with first intrinsic mode function. Hilbert Huang Transform is used for generating instantaneous amplitude and instantaneous frequency feature of real time voltage sag power signal. Outputs of Hilbert Huang Transform is intrinsic mode functions (IMFs), instantaneous frequency (IF), and instantaneous amplitude (IA). Characteristic features are obtained from first IMFs, IF, and IA. The six features—, the mean, standard deviation,skewness, kurtosis of both IF and IA are then calculated. These features are normalized along with the inputs classifiers. The proposed power system monitoring system is able to detect power system voltage sag disturbances and capable of recognize electromagnetic interference component. In this study based on experimental studies, Hilbert Huang Transform based pattern recognition technique was used to investigate power signal to diagnose voltage sag and in power grid. Support Vector Machines and C4.5 Decision Tree were operated and their achievements were matched for precision and CPU timing. According to the analysis, decision tree algorithm without dimensionality reduction produces the best solution. Index Terms— C4.5 decision trees, electromagnetic interference, feature extraction, hilbert huang transform, power quality disturbance, smart grids, support vector machines I. INTRODUCTION Smart grids have been constructed structure where a number of control devices are used to provide reliability, stability and efficiency in the power generation, transmission and distribution. To enhance forecasting faults and risks in addition to ensuring protection against any possible internal and external threats, the new generation, smart grids, will be supplied with communication facilities and real time measurement techniques [1, 2]. The smart grid design is mainly based on restructuring the power industry and optimizing its resources. Smart Grids could optimize transfer capability of transmission and distribution networks to meet the demands for higher quality and more reliable power supply [1, 2]. The main benefits of the Smart Grid technologies include: minimized shutdown of the distributed generation in overload conditions, power quality improvement, improved voltage profile, coordinated restoring of the power system avoid to grid blackout [1,2,3,4]. A. Voltage Sag Voltage sags are short-duration (less than 1 sec) reduction in voltage magnitude. This kind of disturbance is presently one of main power quality problems (Figure 1b.). Momentary increase of current has many origins in power systems such as energizing of transformers, short circuits, earth faults and starting of induction motors [5, 6]. This scientific study is supported by TUBITAK. (Project Number: 114E919) T. Yalcin is with the Ondokuz Mayis University, Department of Electrical & Electronic Engineering, 55139, Samsun, TURKEY (e-mail: turgay.yalcin@omu.edu.tr ). M. Ozdemir is with the Ondokuz Mayis University, Department of Electrical & Electronic Engineering, 55139, Samsun, TURKEY (e-mail: ozdemirm@omu.edu.tr ). B. Electromagnetic Interference (EMI) Electromagnetic interference (EMI), side-effect results of the power conversion and control devices processes, can emerge in a wide frequency range from the basic harmonic and inter-harmonics of the mains frequency. A rise in switching frequencies gives rise to the high energy obstruction, created by the realization of the energy conversion processes, to be shifted in frequency range approximately operated (9 kHz-30 MHz) EMI range. Moreover, a new growing power quality problem especially (2 kHz-150 kHz) threatened the smart grid power quality [7]. EMI normalized voltage signal (L1-N / phase A) generated with arbitrary function generator Tektronix AFG3022C is shown in Figure 1c [2, 6]. Fig. 1. Healthy signal (1a), normalized voltage sag (1b) and EMI normalized voltage signal (1c) (L1-N / phase A) generated with Tektronix AFG3022C 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 -1 0 1 time(sec) A m p lit u d e 1a. Healthy signal 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 -1 0 1 time(sec) A m p lit u d e 1b. Voltage Sag 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 -1 0 1 time(sec) A m p lit u d e 1c. Electromagnetic Interference - Noise mailto:turgay.yalcin@omu.edu.tr mailto:ozdemirm@omu.edu.tr Hilbert Huang Transform method used for recognizing and identifying real time power quality disturbances have described in Section II. II. FEATURE EXTRACTION A. Emprical Mode Decomposition (EMD) The algorithm [8, 9, 10] includes the steps: i. Determine all the extrema of the signal, s(t). ii. Find the upper and lower envelope constructed in step (i). (Interpolation of the extrema analyses with the cubic spline ) iii. Then, find the subtraction signal and the mean function of the upper and lower envelope (mean(t)), dif(t) = s(t)−mean(t). iv. Only when the iteration stops, dif(t) becomes first imf c1(t) ; or else, branch to step (i) change s(t) with dif(t). v. Find the residue signal, res(t) = s(t)−c1(t). vi. Continue the operation from steps (i) to (vi) to attain second IMF, c2(t). Achieve cn(t), continue steps (i) – (vi) after n iterations. The routine is broken when the last imf (residual signal res(t)) is acquired as a monotonic function. This routine called sifting process. Finally, we get residue res(t), gathering of m IMF, from c1(t) to cn(t). The targeted signal can be expressed as: ∑ = += m i i trestcts 1 )()()( (1) we can regard res(t) as cm+1(t) [11, 12,13]. A. EEMD (Ensemble Empirical Mode Decomposition) B. B. Ensemble Emprical Mode Decomposition (EEMD) The EEMD algorithm (Fig. 2.) steps are hearunder: i. Add noise, wn(t), to target signal s1(t). s2(t)=s1(t)+wn(t). ii. Used EMD algorithm for decomposing the final signal s2(t). iii. Continue steps (i) and (ii) till the trial numbers. When new imf combination Cij(t) is succeeded, predict the ensemble mean of the last imf. The aimed output: ∑ = = tn i ijj tctcEEMD 1 )()]([ (2) tn: trial numbers, i: iteration number and j: imf scale [13,14]. Fig. 2. The representation of the EEMD algorithm ∫ +∞ ∞− − = τ τ τ π d t c tv jj )(1 )( (3) cj(t) is real part and vj(t) is imaginary part of an analytic signal zj(t): )()()( tjvtctz jjj += (4) ))(exp()()( tjwtAtz jjj = (5) Amplitude and phase expressed with equation (6) and (7): 22 )()()( tvtctA jjj += (6) ) )( )( arctan()( tc tv t j j j =θ (7) Thus, the instantaneous frequency wj(t) was given by: dt td tw jj )( )( θ = (8) C. Feature Generation: Hilbert-Huang Transform (HHT) HHT [13, 14, 15] enables the real time signal X(t) into the time frequency domain by merging EEMD with the Hilbert transform (Fig. 3.). The Hilbert transform is then implemented for each IMF component Cj generated with sifting process which is explained in Section III.A. Fig. 3. Main steps of the feature generation routine with HHT III. EXPERIMENT SET UP In this part of the study, PQube was installed to acquire measurements, firstly in basic electricity laboratory for one phase (L1-N) records. Secondly it was utilized in computer laboratory for three phases (L1-N, L2-N, L3-N) records which its loads are computers. HHT is used in signal processing part of the study for generation of Instantaneous Amplitude (IA) and Instantaneous Frequency (IF) features. They respectively generated for real time values from PQube one phase in basic electricity laboratory, three phases in computer laboratory for computers. A. Real Time Basic Electricity Laboratory Measurements One phase (L1-N) voltage sag (Fig. 4.) event history which recorded by PQube is shown in Table I. TABLE I EVENT HISTORY (ONE PHASE) Event_Type Voltage Sag Event_Magnitude 60.44% Event_Duration_In_Seconds 0.110 Trigger_Date 2015/11/08 Trigger_Time T 03:05:51.687 Trigger_Channel L1-N Trigger_Threshold 90.0% of nominal Trigger_Sample_Number 257 Samples_Per_Cycle 128 Microseconds_Per_Sample 156.398 Fig. 4. Power Quality Monitor with PQube voltage sag condition of signal (2015/11/08) Fig. 5. illustrates that first component imf1 the noise (lowest magnitude and highest frequency signal) on the line (L1-N). Lower order of imfs means high frequency and oscillation higher order otherwise. Fig. 5. IMFs for a voltage sag signal processed with EEMD -50 0 50 100 150 200 250 300 -500 0 500 time scale Signal -50 0 50 100 150 200 250 300 -50 0 50 time scale imf1 (noise) -50 0 50 100 150 200 250 300 -1000 0 1000 time scale imf2 -50 0 50 100 150 200 250 300 -500 0 500 time scale imf3 -50 0 50 100 150 200 250 300 -200 0 200 time scale imf4 -50 0 50 100 150 200 250 300 -200 0 200 time scale residue Fig. 6. Stockwell Transform (ST) contours The Stockwell Transform (ST) is developed method related with the Gabor Transform (GT) and Wavelet Transform (WT). Several works have used ST for the analysis of PQ disturbance because it allows location in time, real and imaginary components of the spectrum [16, 17, 18, 19]. Fig. 6. shows that ST can produce proper features for detecting voltage sag. Table II shows the main advantages and disadvantages of two signal processing methods (HHT, ST). Table II COMPARISON OF TWO FEATURE EXTRACTION METHOD FOR PQ DISTURBANCES [25, 28] HHT ST Advantages Appropriate for feature extraction of non- linear non-stationary signal, generates perpendicular imfs whereby instantaneous amplitude and phase can be easily assessed Maintain time and frequency representation. Good time-frequency resolution Disadvantages For narrow band conditions is limited, end effects Does not accomplish real-time requirement based on block processing, false harmonics measurement owing to dependency of frequency window width. B. Real time Computer Laboratory measurements For 3 phases (L1-N, L2-N, L3-N) real time processing the first intrinsic mode function is removed with the addition (superposition) of remain components to reconstruct the analyzed signal (Equ. 1.). Respectively, fig. 7. and fig. 8. show that after removing noise component normal and voltage sag cases. Fig. 7. After removing first imf normal condition of signal Fig. 8. After removing first imf voltage sag condition of signal Fig 8. illustrates real time computer laboratory measures that after reconstructing the voltage sag (rate: 86.60%- duration: 0.063 sec) signal, namely the first imf removing from the noisy component. In addition, voltage sag occurred on phase B-C as a result of this case different load types and number of computers on line. This is main vision of this scientific work to identify the fault on active different load types. 0 50 100 150 200 250 -300 -200 -100 0 100 200 300 time scale Am pl itu de PQube Computer LAB. (Normal Conditions) Phase A(L1-N) Phase B(L2-N) Phase C(L3-N) 0 50 100 150 200 250 -300 -200 -100 0 100 200 300 time scale A m p li tu d e PQube Computer LAB. Measurements ("Voltage Sag","86.60%","0.063 sec") Phase A(L1-N) Phase B(L2-N) Phase C(L3-N) Fig. 9. IA corresponding to remove first imf normal condition of signal Fig. 10. IF corresponding to remove first imf normal condition of signal Fig. 11. IA corresponding to remove first imf voltage sag condition of signal The results show explicitly different pattern in Fig. 9.- normal condition as for Fig. 11. – voltage sag condition. Also it is clearly shown in fig. 11. voltage sag on two phases (L2- N, L3-N). This information will be used for evaluating active loads types and risk management of the grid. Fig. 12. IF corresponding to remove first imfs voltage sag condition of signal IF signal can use for separation for two cases but there is end effect problem that have to be solved. This is another future work of the study. When cubic spline fitting is computationally demanding, generates distortions near the end points. This is a technical problem that causes data 0 50 100 150 200 250 305 310 315 320 time scale In st an ta n eo u s A m p li tu d e (I A ) PQube Computer LAB. measurements-Normal condition after removing first IMF(noise) IA-Phase A IA-Phase B IA-Phase C 0 50 100 150 200 250 14 16 18 20 22 24 26 28 time scale In st an ta n eo u s F re q u en cy ( IF ) Normal condition after removing first IMF(noise) IF-Phase A IF-Phase B IF-Phase C End Effect 0 50 100 150 200 250 280 285 290 295 300 305 310 315 320 325 330 time scale In st an ta ne ou s A m pl itu de (I A ) Instantaneous Amplitude after removing first IMF (noise)- Voltage Sag Condition Phase A (L1-N) Phase B (L2-N) Phase C (L3-N) 0 50 100 150 200 250 13 14 15 16 17 18 19 time scale In st an ta ne ou s Fr eq ue nc y (IF ) Instantaneous Frequency after removing first IMF (noise)- Voltage Sag Condition Phase A (L1-N) Phase B (L2-N) Phase C (L3-N) End Effect failures and peaks at the beginning and at the end of the signal. This fault will be investigated on HHT (Fig. 10. - 11.). C. Feature Selection For diagnosis of disturbances, extracted features are produced from firstly EEMD method so as to classify the voltage sags in grid. After reconstruction signal without noisy part, first imf pre-processing stage. Second stage IA and IF) are generated by means of HHT. The statistical analysis and classification for identification power quality disturbances. The following features were extracted: mean, standard deviation, skewness of IA and IF. Selecting appropriate features of voltage sag events are highly crucial for diagnosis of the disturbance. The primary schematic model consists of four steps as shown in Fig. 13. Fig. 13. Schematic model of identification of PQ disturbance IV. PQ DISTURBANCES CLASSIFICATION TECHNIQUES A. Support vector machine Support Vector Machine (SVM) methods, which are developed by Vapnik, whereby statistical learning technique being the basis contributes a novel machine learning method. SVMs are linked supervised learning methods used for classification and regression [20, 21, 25, 28]. B. Decision Trees Decision trees are methods that utilize divide-and- conquer approaches as structure learning by induction [22, 23]. The C4.5 algorithm was developed by Qinlan, contains the generation of a tree whereby a training set, finding the information gain criterion to find the finest attribute/feature to be used at each node. Furthermore, the algorithm applies the post pruning approach to diminish the size of the tree and prohibit over fitting. C4.5 is a technique for approximating discrete-valued functions that is powerful tool to noisy data and suitable for learning distinctive statements [23, 24, 25, 26, 27, 28]. V. PERFORMANCES OF CLASSIFICATION ALGORITHMS AND DISCUSSIONS To figure out the performance of the proposed power quality classification algorithm, a total number of 30 PQube Analyzer real time disturbances data were used. The PQ signals are divided into two categories; 20 of them were used for training and 10 of them were used for testing the proposed algorithm with shuffling the data. In the light of Table III., it is concluded that for sigmoid kernel degree 0.01 with dimensionality reduction with Singular Value Decomposition (SVD) described in [29] is better result in terms of CPU time (3.56 sec), and for polynomial kernel d=3, is also better result CPU time (3.58) in non linear classification SVM. Decision Tree algorithm has the precision of 100% and CPU time of 4.10 sec. Eventually, C4.5 Decision tree based method is the best and gives more proper outcomes than the SVM technique without SVD. (Note: the most proper and robust classifiers for each data set are showed by Red font in Table III). TABLE III PERFORMANCES OF DISTURBANCE DIAGNOSE ALGORITHMS Classifier Precision Time (sec) SVM-Linear 50% 2.35 SVM-poly d=2 100% 3.75 SVM-poly d=2 preprocessing SVD (r=2) 100% 3.59 SVM-poly d=3 90% 1.91 SVM-poly d=3 preprocessing SVD (r=3) 100% 3.58 SVM-poly d=3 preprocessing SVD (r=2) 60% 1.06 SVM-RBF sigma =0.01 50% 0.36 SVM-RBF sigma =0.01 preprocessing SVD (r=3) 100% 3.56 SVM-RBF sigma =1 50% 0.40 SVM-RBF sigma =1 preprocessing SVD (r=3) 50% 0.36 DecisionTree C4.5 100% 4.10 DecisionTree C4.5 preprocessing SVD (r=3) 100% 4.0606 VI. CONCLUSION In this real time analysis, EEMD-HHT signal processing system was used for generation features of different characteristics IA and IF for normal condition and voltage sag http://www.sciencedirect.com/science/article/pii/S0957417407005982#ref_bib7 http://www.sciencedirect.com/science/article/pii/S0957417407005982#ref_bib12 http://www.sciencedirect.com/science/article/pii/S0957417407005982#ref_bib12 cases. The technique reported in this study clearly accomplishes generation of features different for normal – voltage sag cases aiming that identification of smart grid faults. Simulations results have illustrated the capability and validity of the HHT. This study shows that the proposed approach can be easily used for detecting electromagnetic interference on non-stationary signals. Results of the experiments will be conduct for relation on three phases between computer numbers and voltage disturbances for future studies. In PQ Diagnosis part of the study, SVM and Decision Tree (C4.5) were operated and their results were match for precision and CPU time. In consequence of precision and timing criteria, without dimensionality reduction with SVD, SVM-RBF (sigma =0.01) algorithm presented the best solution. Results from the simulations clarify that the proposed method is effective in detecting non- stationary PQ signal. For analyzing the real time power quality disturbance signals and classifying them, Matlab ™ Toolboxes are used for simulations. ACKNOWLEDGEMENT This scientific study is supported by TUBITAK. (Project Number: 114E919) REFERENCES [1] J. Momoh, “Smart Grid: Fundamentals of Design and Analysis”, First Edition, Institute of Electrical and Electronics Engineers, 2012. [2] R. Smolenski, “Conducted Electromagnetic Interference (EMI) in Smart Grids”, Springer -Verlag London, 2012. [3] S. Borlase “Smart Grids: Infrastructure, Technology, and Solutions” , CRC press, 2012. [4] Report to NIST on the Smart Grids Interoperability Standarts Roadmap, EPRI, 2009. [5] M. H. J. Bollen and I. Y. H. Guo, “Signal Processing of Power Quality Disturbances”, New York: Wiley, 2006. [6] P. F. Ribeiro, C.A. Duque, P. M. Silveria, and A. S. Cerqueira, “ Power Systems Signal Processing for Smart Grids”. Chichester, UK: John Wiley & Sons, Inc., 2013. [7] A. McEachern, “Practical Power Quality: An Update, Large Customer Conference”, Power Standards Lab, November 25, 2015. [8] K.-M. Chang,“Arrhythmia ECG Noise Reduction by Ensemble Empirical Mode Decomposition”, Sensors, 10, 2010, pp. 6063 - 6080. [9] N.E. Huang, Z. Shen., S.R. Long, M.L. Wu, H.H. Shih, Q. Zheng, N.C. Yen, C.C. Tung, H.H. Liu, “The empirical mode decomposition and Hilbert spectrum for nonlinear and non-stationary time series analysis”, Proc. Roy. Soc. London A, Vol. 454, 1998, pp. 903–995. [10]Z. Wu, N.E. Huang, “A study of the characteristics of white noise using the Empirical Mode Decomposition method,” Proc. Roy. Soc. London A, 2002. [11] S. Baykut, T. Akgül, S. Ergintav, “EMD – Based Analysis and Denoising of GPS Data”, IEEE 17th Signal Processing and Communications Applications Conference, Antalya, 2009. [12] T.Yalcin, O.Ozgonenel, “Feature vector extraction by using empirical mode decomposition from power quality disturbances”, IEEE SIU, Fethiye, Mugla, 2012. [13] O.Ozgonenel, T. Yalcin, I. Guney, U. Kurt, “A New Classification for Power Quality Events in Distribution System”, Electric Power System Research (EPSR), 95, 2013, pp. 192-199. [14] Z.Wu, N.E. Huang, “Ensemble empirical mode decomposition: a noise- assisted data analysis method”, Adv. Adapt. Data. Anal., 1, 2009, pp.1–41. [15] Z. Wang, Q. Zhu, J. Kiely, R. Luxton, “Hilbert Huang transform impedance measurement data for cellular toxicity monitoring” International Conference on Networking, Sensing and Control, 2009, pp. 767-772. [16] M. Uyar, S.Yildirim, M.T. Gencoglu, “An effective wavelet-based feature extraction method for classification of power quality disturbance signals”, Electr. Power Syst. Res, 78, (10), 2008, pp. 1747–1755. [17] T. Nguyen, Y. Liao, “Power quality disturbance classification utilizing S-transform and binary feature matrix method”, Electr. Power Syst. Res., 79, (4), 2009, pp. 569–575 [18] C.N. Bhende, S. Mishra, B.K. Panigrahi, “Detection and classification of power quality disturbances using S-transform and modular neural network”, Electr. Power Syst. Res., 78, (1), 2008, pp. 122 –128. [19] S. Suja, , J. Jerome, “Pattern recognition of power signal disturbances using S transform and TT transform”, Int. J. Power Energy Syst., 32, (1), 2010, pp. 37–53. [20] B. Biswal, PK. Dash, S. Mishra, “A hybrid ant colony optimization technique for power signal pattern classification”, Expert Syst Appl,38, 2011, pp. 6368–75. [21] K. K. Hoong, S. P. Lam, C. Y. Chung. “An output regulation based unified power quality conditioner with Kalman filters.”, IEEE Trans Ind Electron, 59 (November (11)), 2012, pp. 4248–62. [22] I. H.Witten, E. Frank, “Data Mining: Pratical Machine Learning Tools and Techniques.”, San Mateo, CA, USA: Morgan Kaufmann, 2005. [23] M. T. Hagan, M. B. Menhaj, “Training feedforward networks with the Marquardt algorithm,” IEEE Trans. Neural Netw., , Nov., vol. 5, no. 6, 1994, pp. 989– 993. [24] R. J. Quinlan, “C4.5: Programs for Machine Learning”, San Mateo, CA, USA, Morgan Kaufmann, vol. 1, 1993. [25] S.Mishra, T. Nagwani, “A Review on Detection and Classification Methods for Power Quality Disturbances”, International Journal of Engineering Science and Computing, Volume 6, Issue No. 3, 2016. [26] F. A. S. Borges, R. A. S. Fernandes, I. N. Silva, C. B. S. Silva, “Feature Extraction and Power Quality Disturbances Classification Using Smart Meters Signals”, IEEE Transactions on Industrial Informatics, Vol. 12, No. 2, 2016. [27] Ozgonenel O., Thomas D. W. P., Yalcin T., “Superiority of decision tree classifier on complicated cases for power system protection” , 11th International Conference on Developments in Power Systems Protection, Birmingham, UK, 2012, pp. 134–134. [28] Mahela O. P. , Shaik A. G. , Gupta N., “A critical review of detection and classification of power quality events”, Renewable and Sustainable Energy Reviews, Volume 41, 2015, pp. 495–505. [29] B. Schölkopf, A. Smola, K. Müller, “Nonlinear component analysis as a kernel eigenvalue problem”, Neural Computation, 10, 1998, pp. 1299– 1319. http://ieeexplore.ieee.org/search/searchresult.jsp?searchWithin=%22Authors%22:.QT.Zhiyan%20Wang.QT.&newsearch=true http://ieeexplore.ieee.org/search/searchresult.jsp?searchWithin=%22Authors%22:.QT.Quan%20Zhu.QT.&newsearch=true http://ieeexplore.ieee.org/search/searchresult.jsp?searchWithin=%22Authors%22:.QT.J.%20Kiely.QT.&newsearch=true http://ieeexplore.ieee.org/search/searchresult.jsp?searchWithin=%22Authors%22:.QT.R.%20Luxton.QT.&newsearch=true http://www.sciencedirect.com/science/article/pii/S1364032114007564 http://www.sciencedirect.com/science/article/pii/S1364032114007564 http://www.sciencedirect.com/science/article/pii/S1364032114007564 http://www.sciencedirect.com/science/journal/13640321 http://www.sciencedirect.com/science/journal/13640321 http://www.sciencedirect.com/science/journal/13640321/41/supp/C Turgay Yalçın received the B.Sc. Erciyes University (EU), Kayseri, Turkey in 2006 and M.Sc. degrees in electrical engineering from Ondokuz Mayıs University (OMU), Samsun, Turkey, in 2010 and he is currently pursuing the Ph.D. degree in electrical engineering from Ondokuz Mayıs University (OMU). Currently, he is an research assistant with the Department of Electrical and Electronics Engineering, Ondokuz Mayıs University (OMU), Samsun, Turkey from 2007. His areas of interest are Identification of Power Quality Disturbances, Signal Processing Methods and Machine Learning Algorithms. Muammer Özdemir received the B.Sc. and M.Sc. degrees in electrical engineering from Black Sea Technical University (KTÜ), Trabzon, Turkey, in 1988 and 1991, respectively, and the Ph.D. degree in electrical engineering from The University of Texas at Austin (UT), Austin, TX, USA, in 2002. Currently, he is an Assistant Professor with the Department of Electrical and Electronics Engineering, Ondokuz Mayıs University (OMU), Samsun, Turkey. His areas of interest are power systems harmonics, power quality, and power system analysis. I. INTRODUCTION II. FEATURE EXTRACTION A. Emprical Mode Decomposition (EMD) B. Ensemble Emprical Mode Decomposition (EEMD) III. Experiment Set Up A. Real Time Basic Electricity Laboratory Measurements Table I Event History (one phase) Comparison of two feature extraction method for PQ disturbances [25, 28] B. Real time Computer Laboratory measurements C. Feature Selection IV. PQ DISTURBANCES CLASSIFICATION TECHNIQUES V. Performances of Classification algorithms AND DISCUSSIONS Table III Performances of disturbance diagnose algorithms VI. CONCLUSION References [12] T.Yalcin, O.Ozgonenel, “Feature vector extraction by using empirical mode decomposition from power quality disturbances”, IEEE SIU, Fethiye, Mugla, 2012. [13] O.Ozgonenel, T. Yalcin, I. Guney, U. Kurt, “A New Classification for Power Quality Events in Distribution System”, Electric Power System Research (EPSR), 95, 2013, pp. 192-199. [14] Z.Wu, N.E. Huang, “Ensemble empirical mode decomposition: a noise-assisted data analysis method”, Adv. Adapt. Data. Anal., 1, 2009, pp.1–41.