Neural Network Approach for Fault Detection in Substations

This paper presents a fault detection algorithm for substations. The algorithm was based on neural network approach. The neural network was trained with Levenberg-Marquardt backpropagation algorithm and the training set was formed from the input-output pairs (generator currents, load currents) generated by the substation physical model which was modeled in SIMULINK® by an ideal generator, a three phase transformer and a resistive load. Then under the no-fault condition, the substation neural network model outputs were compared to the substation physical model outputs and maximum errors were computed for each phase(A,B and C). The simulation of the fault detection algorithm consisted of comparing, for each phase, the squared errors between the substation neural network model outputs and the substation physical model outputs; provided that for the fault condition, they would exceed the square of maximum errors previously computed. The simulation results show that the fault detection algorithm is valid and that it can be improved by increasing the size of the training set and by choosing the right neural network architecture.