Power Quality Disturbances Classification Based on Wavelet Compression and Deep Convolutional Neural Network

The power quality disturbances (PQDs) has become an issue of essential importance in the world. The foundation to address the power quality problem is by implementing the PQDs identification and classification technique. This paper presented the 1-dimensional deep convolutional neural network (DCNN) to identify and classify the power quality interferences. The dataset is generated based on the mathematical model of 14 types PQDs which refers to the IEEE-1159 standard. To enhance training time computation, the wavelet compression (WT) method is proposed in the data preprocessing stage. The deep learning architecture is composed of four 1-D convolutional layers, two pooling layers, a dropout layer, a fully connected layer, and a softmax layer. To introduce non-linearity in CNN, this architecture adopts the rectified linear unit (ReLU) function. To demonstrate the DCNN performance, the comparison between the model with the original dataset and the compression dataset is simulated. The experiment result indicates that this approach can successfully predict the PQDs data with more than 99,5 % classification performance, while the computation time improves on the training phase.