An identification method of load harmonic current based on BP neural network

Based on the theory of differential equation model of nonlinear load, an identification method of load harmonic current using BP neural network is proposed. Considering time-varying frequency, the fundamental frequency and voltage on the load can be determined by the windowed discrete Fourier transform and double spectral line interpolation. In order to improve the generalization ability of BP neural network, voltage and current data measured at the connection point of utility grid is checked and Bayesian regularization algorithm is adopted. With the trained BP neural network describing the nonlinear load, the current incented by the fundamental voltage can be obtained. The simulation results demonstrate that the total harmonic distortion of the load current based on BP neural network is almost independent of power capacity and harmonic voltage within the range of utility grid harmonic voltage limits, which is beneficial to the division of harmonic responsibility and harmonic control.