基于BP神经网络的宽频域特定间谐波优化PWM控制方法研究

Research on optimal PWM control method of wide frequency domain specific inter-harmonics based on BP neural network

为解决传统的牛顿迭代法在求解超越方程组计算开关角时依赖给定迭代初值以及无法进行在线特定间谐波控制问题,运用非对称选择性消谐脉宽调制(selective harmonic elimination pulse width modulation, SHEPWM)方法,基于反向传播(back propagation,BP)神经网络,通过对特征谐波的相位进行相移调制实现特定间谐波控制。BP神经网络对原始数据进行训练,在特征谐波控制输入与开关角输出之间构建一种非线性映射关系,即可得到期望的特征谐波幅值、相位与开关角之间的非线性映射关系,不需要求解超越方程组,并且在相移调制过程中,可以利用训练好的BP神经网络实现开关角的在线计算。针对电网中宽频域范围内出现的4类间谐波,在Matlab/Simulink中分别进行基于BP神经网络的特定间谐波控制仿真,以及快速傅里叶变换(fast Fourier transform, FFT)分析,并计算输出波形的加权谐波失真率(weighted total harmonic distortion, WTHD)。与牛顿迭代法的对比结果说明了该方法有较好的特定间谐波控制效果,为特定间谐波的灵活控制提供了可能。

To deal with the problems that the traditional Newton iteration method depends on the given initial value of iteration and cannot carry out online specific inter-harmonic control when it is used to calculate the switching angle to solve the transcendental equations, an asymmetric selective harmonic elimination pulse width modulation(SHEPWM) method is used. Based on the back propagation(BP) neural network, the control of specific inter-harmonics by phase-shifting of the characteristic harmonics is realized. The BP neural network builds a nonlinear mapping relationship between the characteristic harmonic control input and the switching angle output through the training of the original data. That is to say, using this method, the expected nonlinear mapping relationship between the characteristic harmonic amplitude, phase and switching angle can be obtained without solving the transcendental equations. In the process of phase shift modulation, the online calculation of the switching angle can be realized by using the trained BP neural network. Aiming at the four types of interharmonics that appear in the wide frequency domain in the real power grid, the simulation of specific interharmonic control based on BP neural network was carried out in Matlab/Simulink, and the weighted total harmonic distortion(WTHD) and fast Fourier transform(FFT) analysis of the output waveform were calculated. The comparison with Newton iteration method shows that the method has a better control effect of the specific inter-harmonic, and provides a possibility for the flexible control of the specific inter-harmonic.