Switching angles generation for selective harmonic elimination by using artificial neural networks and quasi-newton algorithm

摘要

A hybrid method based on artificial neural networks (ANNs) and Quasi-Newton algorithm is proposed to generate the switching angles for selective harmonic elimination (SHE), which makes a compromise among the memory consumption, executing efficiency and the solution precision. Unlike the other ANNs based methods which use ANNs to directly give the final switching angles, this hybrid method just uses ANNs to give the initial values, which lowers the precision requirement on training the ANNs, so, the number of the neurons can be reduced and less on-chip memories are required. Then, the Quasi-Newton algorithm is used to solve the exact switching angles from the initial values given by the ANNs, which guarantees the solving efficiency and the solution precision. The case of 11-level staircase modulated converter is studied by using the single-layer back-propagation (BP) neural networks. The trained neural networks have only 9 neurons in the hidden layer and the output initial values can meet the convergent requirement of the Quasi-Newton algorithm in the full range of modulation index. The total executing time is about 70ms on a STM32F407 microcontroller, as the executed code is automatically generated by MATLAB, the executing time could be further reduced if the code is manual optimized. Experiment results are also shown to verify the correctness of the switching angles generated by the proposed hybrid method.