Reinforcement Learning based Control for Non-Isolated DC-DC Converters

Abstract

In this technical report, we will review the application of Reinforcement Learning (RL) in electrical systems and apply the algorithm to control the non-isolated dc-dc converters i.e., Buck, Boost and Buck-Boost. The RL algorithm takes action by updating the action value function, also called the Q-function. The reward function is calibrated to choose the best possible action on the basis of its effort in minimizing the error state i.e., the difference between measurement and reference. Mathematical derivations will be supported by simulation results to prove the application of this theory.