Comparing Support Vector Machine and Artificial Neural Networks Based Model Predictive Control in Power Electronics

There is increasing interest in using the model predictive control (MPC) for power electronics converters. The computational complexity prevents it from being implemented in the real-time digital signal processor (DSP) with limited computation resources. Recently, machine learning (ML) based MPCs are used to reduce the computation of the MPC. However, the majority ML-MPCs are based on the artificial neural networks (ANN), the O(2 ⁿ ) computation complexity still prevents it being implemented in the DSP. This paper proposes a linear support vector machine MPC (LSVM-MPC) with O (1) computational complexity such that it can be implemented like a PID controller. Besides, more historical data is used to increase its steady-state performance. The experiment is carried out in a three-phase inverter with L-C filter. The results show that the linear SVM-MPC can achieve better performance than the ANN regardless in linear resistive or nonlinear RCD load, e.g., in the linear load, with the linear SVM-MPC, 1.13% THD is achieved, which is lower than 1.58% with the ANN.