Fault Analysis in Electrical System using Machine Learning Algorithms

Abstract

This research provides an overview of the prediction accuracy of different classification-based machine learning algorithms from a ‘Simulink based 3-phase Electrical System Model’. The model consists of a three-phase source and two relay bus bars connecting the ends of two subsystems. The data generation for faults was embedded by connecting a three-phase fault block between two subsystems. Later, the whole simulation was explained by interchanging the parameters among phase A, B, C and Ground. Each of the simulation placed between 0 to 1 second. The data table exhibits more than 16,000 samples across voltages ranging from -0.1023 V to 1.9776 V and a current ranging from -0.13813 A to 12.53622 for each of the faulty phase. More than 70% of the data was used for model training later and rest of the 30% raw data was planned for prediction. For classification, KNN, SVM, LR, DT, Gradient Boosting, Random Forest and MLP Classifiers algorithms were used for comparison. After several Preprocessing, ‘Random Forest’ algorithm comes with the highest accuracy of 0.9477 or 94.77%. Thus, the prediction can help the electrical engineers to automate the process of fault detection with the help of machine learning.