Abstract: “Neural networks have been proved as an important and useful tool for solving a wide variety of practical and real-world problems. Huge research in this field alleviated in understanding and finding new and effective methods to address different problems. However, selection of apposite combination of training and transfer function for a particular problem is a cumbersome task. But, this can be ascertained through research experiences and outcomes. The objective of this work is to compare the performances of three transfer functions in tandem with fourteen training functions used for backpropagation training of neural network for recognition of power quality (PQ) disturbance signatures. The comparison is shown on the basis of Lowest MSE, number of epochs, convergence time, and accuracy. It is shown that among three transfer functions namely “logsig”, “purelin”, and “tansig”; the overall performance of “tansig” was superior and the accuracy of BR training function was 100 % with all the t”

Keywords: Artificial neural network, feedforward neural network, power quality, recognition, signature