A Harmonics Analysis Method Based on Triangular Neural Network

Abstract

Aiming at fast and effectively evaluating harmonics in the power system, a triangular neural network is constructed, of which the hidden neurons are activated with triangular functions. Based on gradient descent method, the learning rules (i.e., weights-iterative-formula) for the constructed neural network are derived. Then global-convergence of the weights-iterative-formula is proved. As the results, a weights-direct-determination method is achieved, which could obtain the optimal weights of such a neural network in one step by using pseudo-inverse. Furthermore, several numerical tests have been conducted to apply this method to some harmonics models. The simulation results substantiate this method can be used to fast and precisely evaluate the harmonic components.