Adaptive electric vehicle battery charging with multi-objective reinforcement learning

Abstract

Fast and safe battery charging remains a critical barrier to large-scale EV adoption, as high charging rates accelerate battery degradation and increase thermal risk. Traditional control strategies often rely on static heuristics or single-objective optimization, limiting their ability to manage trade-offs between speed, safety, and battery longevity. This work proposes a deep multi-objective reinforcement learning (MORL) framework for optimal EV battery charging. The proposed agent learns policies that dynamically balance competing objectives—such as minimizing charge time and thermal stress—based on user-defined preferences. Unlike scalar reward methods, MORL captures trade-offs explicitly and adapts charging behavior to context. Experimental results show the policy’s adaptability: faster charging is achieved when temperature constraints are relaxed, while more conservative profiles emerge when battery longevity is prioritized. This highlights the potential of MORL to enhance both the safety and efficiency of EV charging.