Optimization study of an R290 electric vehicle thermal management system using a synergistic deep reinforcement learning and genetic algorithm

High-performance thermal management system (TMS) using environmentally benign refrigerants is critical to the safe, efficient operation of electric vehicles (EVs) and to reducing environmental impact. An indirect TMS based on an R290 vapor-injection (VPI) heat pump was developed in this study, and a simulation platform for the TMS was established based on AMESim software to evaluate system performance. We further propose a hybrid multi-objective optimization (MOO) strategy that couples deep reinforcement learning with non-dominated sorting genetic algorithm II (DRL-NSGA) to determine the optimal operating parameters for system performance and energy efficiency. A theoretical analysis based on hypervolume drift confirms that the algorithm's "warm-start" strategy accelerates convergence and enhances solution diversity. The results demonstrate that, compared to conventional MOO methods, DRL-NSGA produces Pareto fronts with broader coverage and higher solution quality while reducing computational time by more than 60%. Under the optimized settings, the system's coefficient of performance (COP) was enhanced by up to 7.7%, with cooling and heating capacities increasing by up to 7.2% and 5%, respectively. Furthermore, the time required to reach the target outlet air temperature was significantly reduced by 132 s under cooling and 429 s under heating conditions. The proposed synergistic optimization algorithm offers an effective and efficient solution to MOO in EV thermal management.