Predictive Model for Classification of Power System Faults using Machine Learning

Power System is a combination of electric power generation, transmission, distribution and utilization systems. In brief, power system is the heart of any electrical system. In an electric power system, a fault or fault current is any abnormal electric current. As a consequence of such fault, the entire system may damage and eventually collapse. The aim of this work is to automatically classify the faults into one of the eleven faulty classes, which includes both balanced and unbalanced faults. The dataset of generated fault in overhead transmission lines is synthetic, which consists of 11 different faults for 100 kilometers. The simulation is done using MATLAB/Simulink software model. The task of classification of faults is implemented using supervised machine learning algorithms in Python and scikit-learn. Comparison is made using three commonly used classification algorithms - Decision Tree (DT), K-Nearest Neighbor (KNN), Support Vector Machine (SVM). SVM performed excellent giving a performance with 91.6% test accuracy for the generated dataset. The predictive model will thus make the system more intelligent in bringing up reliable power supply.