A Reinforcement Learning Compensated Feedback Linearization Controller for Voltage Regulation of Constant Power Loads

Feedback linearization controllers are widely used for boost converters that regulate the voltage of the power systems integrated by constant power loads (CPL). However, state-of-the-art feedback linearization controllers have two drawbacks. First, feedback linearization controllers are usually designed based on the approximated model of the boost converters, not the detailed model. Second, the feedback linearization controllers are actually model-based. When there are mismatched parameters in the boost converters, the performance of the feedback linearization controller will deteriorate. To handle the model approximation and mismatched parameters, this paper proposes a reinforcement learning compensated feedback linearization (RL-FL) controller. The proposed RL-FL controller utilizes the deep deterministic policy gradient (DDPG) to compensate for the control actions obtained by the feedback linearization (FL) controller to enhance the performance of voltage regulation. Afterwards, a case is studied to verify the enhanced performance of the proposed RL-FL controller compared with the FL controller.