Reinforcement Learning-based Control of a Buck Converter: A Comparative Study of DQN and DDPG Algorithms

Reinforcement learning (RL) has emerged as a promising approach for controlling power electronics systems due to its ability to control nonlinear systems without accurate models, and handle unknown changes. This paper presents a comparative study of two widely used deep RL algorithms, Deep Deterministic Policy Gradient (DDPG) and Deep Q-Network (DQN), for controlling the output voltage of DC/DC buck converters. DDPG and DQN are tailored to handle continuous and discrete environments, respectively. The performance of these algorithms is evaluated through simulations with steady-state error, overshoot, settling time, and training time serving as evaluation metrics. The simulation results demonstrate that both DDPG and DQN show excellent performance in controlling power converters, with DDPG generally performing better than DQN in systems with unpredictable changes.