Supervised Learning Model Predictive Control Trained by ABC Algorithm for Common-Mode Voltage Suppression in NPC Inverter

Training the weighting factors of model predictive control in multiobjective problems is a time consuming and sophisticated process. In this article, conventional model predictive control (CMPC) has been developed as supervised learning model predictive control (SLMPC) to cancel common-mode voltage (CMV) in a three-phase neutral-point-clamped (NPC) inverter, while other control objectives are desirably tracked. SLMPC is accurately and quickly trained through the artificial bee colony (ABC) algorithm to optimize the controller weighting factors. Using the optimized weighting factors, transient response is minimized and CMV is surpassed. After training the weighting factors, SLMPC containing the optimized waiting factors is applied to the three-phase NPC inverter without considering the ABC algorithm in the control loop. By applying the optimized weighting factors to the cost function, SLMPC has been evaluated under several experimental and simulation tests to show that desired control objectives, particularly CMV suppression, have been attained. The proposed training process can be generalized and used for MPC cost functions with more control objectives to obtain the best possible performance.