Reinforcement Learning Based Weighting Factor Design of Model Predictive Control for Power Electronic Converters

Weighting factor design is one of the challenges for finite-set model predictive control (FS-MPC) controlled power electronic converters, which plays an important role in the balance of control objectives in the cost function to achieve desired performance. This paper investigates the application of reinforcement learning algorithm for the weighting factor design for FS-MPC regulated voltage source converter in uninterrupted power supply (UPS) system. The deep deterministic policy gradient (DDPG) agent is employed to learn the optimal weighting factor design policy. The reinforcement learning (RL) agent is trained in the system and the weighting factor is optimized based on reward calculation with the interactions between the agent and environment. The key performance metric, total harmonic distortion (THD), is incorporated in the reward function. Effectiveness of the proposed reinforcement learning based weighting factor design method is validated by simulations.