基于神经网络的电力系统高精度频率谐波分析

Neural Network Based High Precision Power System Frequency Harmonic Analysis

加窗插值FFT算法是电力谐波分析常用的高精度算法，但在严重非同步采样情况下，其谐波分析精度有限。该文提出一种基于神经网络的高精度电力系统频率谐波分析算法。采样频率不能与实际基波频率同步时，该算法通过对与基波频率、谐波幅值及相位等相关参数进行更新，当神经网络收敛时，可以获得高精度的谐波分析结果。仿真结果表明，当基波频率在40-60Hz范围变化时，电力系统基波频率、基波和谐波幅值和相位的分析精度超过99．999999999％。

Window and interpolation algorithm is now widely employed to estimate power harmonics parameters, but its precision is limited in extreme asynchronous sampling cases. In this paper, a neural network algorithm is proposed for accurate estimation from periodic signals in power systems. It is aimed at the system in which the sampling frequency cannot be locked on the fundamental frequency. This new algorithm updates the parameters related to fundamental frequency, harmonic amplitude and phase, and when the algorithm convergent, accurate harmonic analysis results can be obtained. The simulating results show that the estimated fundamental frequency, harmonic amplitude and phase can be measured at accuracy of 99.999999999% with actual fundamental frequency varying from 40Hz to 60 Hz.