基于神经网络和对称秩的特定谐波消除开关角生成算法

利用查表法实现三相级联11电平逆变器特定谐波消除面临存储开关角多、占用内存大的问题；而利用小规模神经网络拟合则面临生成开关角精度低的问题。针对这些问题，文中提出基于小规模神经网络和对称秩迭代相结合的特定谐波消除开关角生成算法。首先，利用小规模神经网络预测开关角的迭代初值，降低对神经网络训练精度的要求；其次，利用对称秩算法迭代开关角初值得到精确解。与查表法和Hopfield神经网络算法相比，所提算法可以减少数据存储空间，同时生成精度较高的开关角。仿真和实验结果表明:所生成的开关角能有效地消除5次、7次、11次、13次谐波，同时保证了期望的基波幅值。

Switching Angle Generation Algorithm for Selective Harmonic Elimination Based on Neural Network and Symmetric Rank

When the look-up table method is used to realize the selective harmonic elimination of three-phase cascaded eleven-level inverter, the problems of massive storage switching angles and memory occupation are obvious. And there is the problem of low generation accuracy when the small-scale neural network is used to fit switching angles. To address these problems, this paper proposes a switching angle generation algorithm for selective harmonic elimination based on small-scale neural network and symmetric rank iteration. Firstly, the small-scale neural network is used to predict the switching angles as iterative initial values, which can reduce the requirement of training accuracy of neural network. Secondly, the symmetric rank algorithm is used to iterate the initial values of switching angles, which can obtain precise solutions. Compared with the look-up table method and the Hopfield neural network algorithm, the proposed algorithm can reduce data storage space and generate high-precision switching angles. The results of simulation and experiment show that the generated switching angles can effectively eliminate 5th, 7th, 11th, and 13th harmonics while retaining the desired fundamental amplitude.