Power quality disturbance classification

Abstract

The utilization of nonlinear loads, to facilitate life easier with the technical advancements increases the power quality issues in the electrical power system. Hence to protect the system, it has become necessary to mitigate the Power Quality Disturbances (PQD). To find an effective and efficient method of mitigation it is important to identify and categorize the newlinePQD issue properly. It is therefore logical to develop techniques for automatic disturbance identification, which are applied directly or through feature extraction and pattern recognition. This thesis aims and presents algorithms for detecting and classifying PQDs with Neural Pattern Recognition (NPR) technique, Machine Learning (ML) Techniques, and Deep Learning (DL). The classification algorithms are trained and tested with various power quality issues that occur frequently in a distribution system and found to be effective. newlineObjectives of the thesis work are 1. To detect and categorize the PQD events with the NPR technique. To decompose the PQD signals through DWT and HT, to generate the input and target vectors, to train the NPR network with feature vectors, to test the trained network with confusion matrix and ROC and hence to classify the PQD events. 2. To classify the PQD events with various Machine Learning techniques. To decompose the PQD signals through DWT, to extract the features, to train the ML algorithm with the extracted features newlinethrough Support Vector Machine (SVM), K Nearest Neighbor (kNN) and Decision Tree(DT) and hence to classify the PQD events. newline newline newline