Electric vehicle battery charging with safe-RL

• Autores: Trimboli, Maximiliano; Avila, Luis; Antonelli, Nicolás
• Año de publicación: 2023
• Idioma: inglés
• Tipo de recurso: documento de conferencia
• Materia: Ciencias Informáticas; Safe-RL; State of Charge; Battery aging; Variability
• Institución: Universidad Nacional de La Plata

To become the standard power supply for electric vehicles(EVs), Li-ion batteries need balanced current profiles in order to avoidundesirable electrochemical reactions and excessive charging times. Inthis work, we propose a safe exploration deep reinforcement learning(SDRL) approach in order to determine optimal charging profiles undervariable operating conditions. One of the main advantages of reinforce-ment learning (RL) techniques is that they can learn from interactionwith the real or simulated system while incorporating the nonlinear-ity and uncertainty derived from fluctuating environmental conditions.However, since RL techniques have to explore undesirable states beforeobtaining an optimal policy, no safety guarantees are provided. The pro-posed approach aims at maintaining zero constraint violations through-out the learning process by incorporating a safety layer that corrects theaction if a constraint is likely to be violated. Tests performed on theequivalent circuit of a li-ion battery under variability conditions showearly results where SDRL is able to find safe policies while consideringa trade-off between the charging speed and the battery lifespan.
Sociedad Argentina de Informática e Investigación Operativa