Evaluation of an Intelligent Controller for Electric Vehicle Charging with Vienna Rectifier Topology

The Vienna rectifier is a widely used topology in a variety of industrial applications, including electric vehicle (EV) charging. Although conventional control approaches exist, the nonlinear and highly dynamic nature of the Vienna rectifier as well as many existing uncertainties within the system hinder the development of a performant controller based on classic control theory. This paper proposes a model-free controller based on reinforcement learning (RL) to optimize the performance of the DC-link voltage control loop with concurrent regulation of the power factor. We developed a state-space model of the Vienna rectifier in MATLAB/Simulink and a classical proportional-integral (PI) controller for quantitative comparison. The RL controller's ability to handle system nonlinearities and component uncertainties, such as inductor and capacitor tolerances, enables it to outperform classical control strategies. Furthermore, this paper benchmarks multiple RL-based algorithms including deep deterministic policy gradient (DDPG) as well as twin delayed DDPG (TD3) in the voltage control loop. The results demonstrate that TD3 can reduce 2.5 percent voltage ripple and has a shorter settling time of 0.75 s at the cost of higher computational requirements.