The increasing importance of electric vehicles requires addressing challenges
related to fast charging, safety, and battery range. Thermal management ensures
safety, prolongs battery life, and enables extremely fast charging. In this
regard, this article proposes a novel battery thermal management system (BTMS)
optimization approach based on a model-free deep reinforcement learning (RL) for
a battery pack of an electric vehicle under extreme fast-charging conditions
considering the detailed dynamics of vehicle-level BTMS. The objective of the
proposed approach seeks to minimize the battery degradation and power
consumption of the underlying BTMS. In this respect, the dynamic equations of
the thermal system model are constructed considering the air-conditioning
refrigerant loop and indirect battery liquid cooling loop. Further, the proposed
methodology is implemented on a battery pack, and the results are compared with
those of model predictive control (MPC) and proportion–integral–derivative (PID)
as representatives of optimal control and tracking control baseline strategies,
respectively. It is shown that if a perfect BTMS model is at the disposal of
MPC, MPC performance can be as good as that of the proposed RL. However, the
proposed RL algorithm is 48 times faster than MPC during testing. The
superiority of the proposed deep RL over MPC is attributed to its model-free
nature. Lastly, the RL outperformance is shown over the PID controller in terms
of both battery degradation and BTMS power consumption, where the former is
improved by up to 1.05% whereas the latter is improved by up to 43.68%.
