Economical Thermal Energy Management for Hybrid Electric Vehicles Based on Multiple Agents Deep Reinforcement Learning

Hybrid electric vehicles (HEVs) encounter the challenge of restricted versatility in harsh weather conditions, as the thermal management system (TMS) consumes noteworthy energy to control battery and cabin temperature. This article proposes an economical thermal energy management system (eTEMS) and achieves multiobjective optimal control based on a multiple agents deep reinforcement learning (DRL) algorithm. First, a control-oriented coupled model of a hybrid powertrain and TMS is designed to minimize energy loss. Second, the multiple agents deep deterministic policy gradient (DDPG) is formulated to attain cooperative optimization of fuel economy, battery lifespan preservation, and cabin thermal comfort. Moreover, centralized and decentralized training architecture is tailored to enhance training efficiency and ensure optimal performance. Finally, the real-time performance of the proposed strategy is verified by hardware-in-the-loop (HIL). The results of the HIL experiment indicate that the proposed strategy can reduce the vehicle driving costs in cold and hot environments by 13.28% and 7.1% respectively, while ensuring battery life and cabin thermal comfort. The real-time performance of the proposed strategy is maintained at the millisecond level in the HIL experiment, providing a comprehensive energy-saving solution for HEVs under diverse weather conditions.