SELECTION OF PROPER ACTIVATION FUNCTIONS IN BACK-PROPAGATION NEURAL NETWORKS ALGORITHM FOR IDENTIFYING THE PHASE WITH FAULT APPEARANCE IN TRANSFORMER WINDINGS

摘要

This paper presents an algorithm based on a combination of Discrete Wavelet Transforms and back-propagation neural networks for identifying the types of fault including the phase with fault appearance of a two-winding three-phase power transformer. Fault conditions of the transformer are simulated using ATP/EMTP in order to obtain current signals. The training process for the neural network and fault diagnosis decision are implemented using toolboxes on MATLAB. Various cases and fault types based on Thailand electricity transmission and distribution systems are studied to verify the validity of the algorithm. Various activation functions in each hidden layer and the output layer are compared in order to select the best activation function for identifying the types of internal fault of the transformer winding. It is found that average accuracy obtained from hyperbolic tangent - hyperbolic tangent - linear activation function gives satisfactory accuracy, and will be particularly useful in the development of a modern differential relay.