Double Deep Q Networks Reinforcement Learning-Based Dynamic Weighting Factor of FCS-MPC for Multilevel Inverters

Finite control set model predictive control (FCS-MPC) is one of the widely used control techniques for multilevel inverters (MLI) where multiple control objectives can be included in the cost function to be minimized at each time step. The priorities of the control objectives are determined by weighting factors that can either be configured with constant values that are optimal across a spectrum of operating conditions or dynamically adjusted based on the variations in the operating conditions through an auto-tuning mechanism. This paper proposes a reinforcement learning (RL) based algorithm to auto-tune weighting factors in FCS-MPC for multilevel inverters (MLIs) under varying operating conditions. The designed auto-tuning agent is trained and tested on single-phase grid-connected 9-level crossover switches cell (CSC9) MLI. Simulation results for the CSC9 inverter are curried out to demonstrate the efficacy of the proposed design in decreasing the total harmonic distortion (THD) of the generated current and minimizing capacitor voltage error. Comparison between the proposed solution and the cases with constant weighting factors and alternative auto-tuning methods is provided.