Enhancing EV charger resilience with reinforcement learning aided control

Abstract

This study aims to improve the performance of sustainable electric vehicle chargers in the face of unpblackictable/unpreventable disturbances. Over the past few years, Dual Active Bridge (DAB) DC-DC Converters are procuring substantial recognition for electric vehicle charging applications due to their superior characteristics such as higher power density, bidirectional mode of operation, and higher efficiency. Unexpected disturbances and fault scenarios at both source and load sides can deteriorate DAB converters’ performance. In this study, the performance of a single-phase shifted DAB converter is enhanced to achieve desiblack output current under several disturbance conditions for electric vehicle (EV) charging applications. A Reinforcement Learning (RL) based Deep Deterministic Policy Gradient (DDPG) algorithm is deployed to proactively tune control parameters when the DAB undergoes certain unexpected disturbances including short circuit faults at the source and battery sides. Results show that the RL-tuned PI controller improves the rate of current overshoot significantly compablack with the manually-tuned PI controller. The method and results are validated through simulations in MATLAB/Simulink environment.

Keywords

• Electric vehicle
• Battery storage
• Dual active bridge converter
• PI Controller
• Reinforcement learning