PQ disturbance detection and classification combining advanced signal processing and machine learning tools

This chapter presents a novel approach by combining advanced signal processing (ASP) and machine learning tools (MLTs) for detecting and classifying power quality disturbances (PQDs) combining ASP and MLTs. Among many ASP techniques, this chapter selects two techniques, namely wavelet transform and Stockwell transform for their comparative advantages over others for extracting useful features from PQD recorded signals. Then, the extracted features are fetched for different configurations of artificial neural networks (ANNs) by training and their performance is tested to detect and classify PQD signals. In the proposed research, the key parameters are varied, including the number of neurons in the hidden layer and training algorithms to achieve the best-suited ASP-inspired ANN model. Simulation results suggest the ANN models with Levenberg–Marquardt (LM) training algorithms achieve better generalization performance over the other models. The ANN models with LM training algorithms also exhibit excellent accuracy and robustness against different levels of measurement noises.