A Linear Support Vector Regression-Based Model Predictive Control with Repetitive and PI Elements for a Three-Phase Inverter

The conventional machine learning (ML)-based model predictive control (MPC) methods have the same inputs as the conventional MPC and, therefore, cannot outperform the conventional MPC which is optimized in every sampling period, i.e., online MPC. However, the use of historical data in a proper manner may help ML-based MPCs outperform online MPC. This paper proposes a linear support vector machine (LSVR) MPC for three-phase inverters in which historical errors are selected so that the LSVR-MPC can be interpreted as a repetitive and proportional integral (RPI) controller. The simulation results of a three-phase inverter show that the RPI-LSVR-MPC outperforms the conventional MPC in terms of tracking errors and total harmonics distortion (THD) of the inverter output voltage. The experiments carried out in a dSPACE MicroBox further validate the effectiveness of the LSVR-MPC in reducing the voltage THD from 1.31% when using the online MPC to 0.87%. Furthermore, since the overall control algorithm is linear, it can be easily implemented like a PID controller.