Using Convolutional Neural Network for Fault Identification and Location in a Power System

Abstract

In this work, Convolutional Neural Network (CNN) is applied to the fault identification and location in a small power system. The phasor measurement units are assumed installation on all load buses to measure the bus voltage magnitudes and the active powers of the transmission lines in the time domain. Those records of the measured values are used as the input data of CNN, and these records are split into the training and testing groups. A4-machine 8-bus power system is used to evaluate the performances of CNN in this work. Three fault types of three-phase fault, single line to ground fault, and line break fault are simulated in the power system for evaluating the CNN-based identifying ability of the fault type. Two major fault scenarios of the single fault and two different faults occurrence at the same time are simulated for training CNN model. The accuracy rate of the fault identification is used to evaluate the performance of the proposed method in this paper. The simulation results have shown that the accuracy rate is 100% when all records are in the training group. The accuracy rate for the other cases is above 80%.