Fault Classification by Using Various Neural Network Architectures Based on PSCAD

Abstract

Transmission lines are the most sensitive part of power system network and holds the highest percentage of fault occurrence. A transmission line model is created in this paper using PSCAD software which simulates all types of faults. A useful alternative to conventional relays based on artificial neural network is proposed in this study. Neural networks have strong capability to learn complex relationships with the help of data presented to it. Here, data generated by PSCAD model is provided to the neural network, which are sets of three currents and voltages. First network of 6-10-4 configuration is created, based on the default setting, gives mean squared error of 0.051 and a correlation of 0.88. Change in performance was observed in varying influential parameters of neural network, which includes hidden neurons, layers and transfer function. Final model having configuration 6-17-6-4 was selected, which reduces the error to 0.017 and the overall correlation rose up to approximately 0.97. The whole neural network process was carried out on MATLAB® software.