Power system harmonic analysis based on RPROP ANNs

For improving the speed and accuracy of harmonic analysis, a harmonic analysis method based on RPROP neural network is proposed. Hanning-windowed interpolation harmonic analysis algorithm is used to obtain initial weight and bias values of ANNs(Artificial Neural Networks) and the network takes RPROP algorithm as a training algorithm derived number. Different from the BP(BackPropagation) algorithm, the algorithm adjusts the parameters of ANNs by sign information of first-order partial derived number, which avoids the influence of amplitude information of first-order partial derived number which is useless for parameters adjustment. In the meantime, the algorithm does not have the problem of parameters selection. Therefore, the convergence speed, accuracy and real-time performance of power system harmonic analysis can be improved. Finally, the variable learn rate and momentum term BP ANNs and RPROP ANNs are used to analyze signal, and the comparison of the results verifies the conclusion.