Power Quality Disturbances Classification with Deep Learning Approach

Multiple energy systems are being integrated with the modern power gird to fulfill the growing power demand. But this gives rise to the risk of degradation in the power quality in various stages like generation, transmission, and distribution. The very first step to resolve the risk is to perceive and categorize the PQDs (Power Quality Disturbances). The traditional approach divides this process into various independent stages like analysis of the signal, extraction of various features, optimal feature selection, and finally classification according to the selected features. However, the signal analysis method has its inherent defects and the manual selection of feature is very tedious and imprecise. So the traditional models have lower classification efficiency and poor noise immunity. In practical cases, with higher noise level the traditional model could be greatly affected by the external conditions while classifying the disturbances. In this paper, a convolutional neural network (CNN) model based on deep learning methodology is proposed to replace the traditional models. Total 16 classes of disturbances including combined disturbances are simulated using IEEE-1159 standard. Using continuous wavelet transform (CWT), the coefficient matrix of the signals is generated. Then the signals are converted to image format using the feature matrix. Then the converted images are given as input to the CNN model for automatic classification of PQDs. Using MATLAB a large number of signals with different PQDs are synthesized with various levels of noise as 20dB, 30dB, and 40dB to include the various external conditions. Then the signals are classified using the proposed model without going through the traditional steps. Finally, the comparison results of the proposed model with several traditional models are presented. The suggested model is found as an accurate disturbance classifier and remained unaffected by the external noise conditions.