POWER QUALITY DISTURBANCE DETECTION, CLASSIFICATION AND MITIGATION USING WAVELET-NEURAL NETWORK: CASE OF DIRE DAWA ELECTRIC POWER DISTRIBUTION NETWORK

Abstract

Power quality has become a vital issue recently due to the expansion of electrical load consumption and the growth of sensitive device usage in power systems. To maintain power quality and to ensure its reliability, power quality disturbances must be detected and classified correctly and precisely. Electric power disturbance is going to be a day-to-day phenomenon in Ethiopia. Among different Ethiopia regions, Dire Dawa is one part that is facing such phenomena. Thus, a modern power quality disturbance detection, classification and mitigation technique for Ethiopia in Dire Dawa electric power distribution network is necessary. Specifically, in this study wavelet-neural network is deployed to the Ethiopian context using collected data from Ethiopian Electric Utility. This technique includes feature extraction and detection, classification, and finally mitigation power quality disturbances. Eight signals, one normal signal, and seven power quality disturbances such as voltage sag, swell, interruption, harmonics, sag with harmonics, swell with harmonics, and interruption with harmonics are included in the study. In feature extraction and detection stages, the voltage signal of these power quality disturbances is simulated and generated and then changed into relevant data using wavelet transform. The feature extraction is carried out with energy features, approximation, and detail coefficients to reduce the dimensions of input data. In the classification stage, extracted features obtained from approximation and detail coefficients waveforms, serve as input data for training and testing of the classification stage. 578 samples with 14 features of eight signal classes are used as input to the classifier. The wavelet-neural network results show that the algorithm has high performance with minimum and overall accuracies of 98% and 99.32%. The classifier models like medium tree, coarse tree, and bagged tree are used to plot confusion matrix showing the accuracy and performance of artificial neural network. Finally, Unified Power Quality Conditioner is used as a mitigating mechanism to remove voltage-related power quality problems and improve power quality. The unified power quality conditioner improved the power quality by balancing the load voltage to its nominal value when either of the disturbances such as voltage sag, swell, or interruption occurred in the system.