Solution for Selective Harmonic Elimination in Asymmetric Multilevel Inverter Based on Stochastic Configuration Network and Levenberg-Marquardt Algorithm

Abstract

A hybrid method based on stochastic configuration networks (SCNs) and Levenberg-Marquardt (LM) algorithm is proposed to generate switching angles of selective harmonic elimination for asymmetric multilevel inverter, which makes a compromise among the optimal neurons in neural networks, executing efficiency and the solution precision. Unlike the other artificial neural network (ANN) based methods which use ANN to directly give the final switching angles, this hybrid method just uses SCNs to give switching angles initial values, which greatly lowers the training precision requirement, and requires less on-chip memories for weights and biases of neural networks. Then LM algorithm is used to solve the exact switching angles from the initial values given by SCNs, which guarantees the solving efficiency and the switching angles precision. The case of 7-level asymmetric multilevel inverter with 3 groups of unequal dc-link voltages in the full range of modulation indexes is studied. Compared to the high dimensional look-up table method, data storage space of the hybrid method is decreased by 92%, and the errors of solutions for switching angles are 1e-2 degrees. The results of simulation illustrate SHE switching angles generated by proposed method can effectively eliminate 5th-, 7th-order harmonics while retaining the desired fundamental.