Data-Driven Switching Control Technique based on Deep Reinforcement Learning for Packed E-Cell as Smart EV Charger

Among Hybrid Multilevel Rectifiers (HMRs), Packed E-Cell appeared as an interesting topology due to the generation of nine-level voltage with minimum active/passive devices. But appropriate control design of PEC rectifier is vital demand to keep capacitors voltages well-regulated even under unbalanced/variable dc loads. Therefore, the Backstepping Control (BSC) strategy is developed to control a nine-level Packed E-Cell (PEC9) rectifier to be used as a smart EV charger. Proximal policy optimization (PPO) with Actor and Critic deep neural networks (ADNNs and CDNNs) are trained to adjust the BSC controller where the PEC9 rectifier can intelligently deal with asymmetrical/symmetrical dc loads. By maximizing a reward function, the PPO agent tries to find the optimal policy to design the control coefficients of BSC with the aim of regulating the PEC9 capacitor’s voltages. The developed BSC based on the PPO tuner is validated using Hardware-in-the-loop (HiL) and experimental implementation of the PEC9 rectifier to assess the performance of the proposed control scheme.