Reinforcement Learning-based Controller for Thermal Management System of Electric Vehicles
Wan-Sik Choi, Jae Woong Kim, Changsun Ahn, Juhui Gim
Abstract
The thermal management system in electric vehicles (EV) becomes significant because the performance of the system is highly correlated with the driving range, reliability, and safety of the electric vehicles. Therefore, a controller of the thermal system should be designed to minimize the tracking error and power consumption while satisfying constraints. In this study, a reinforcement learning (RL)-based controller is proposed. This paper presents the selection of states and design of the reward function of RL for the thermal management system of EV. The controller is trained by the sequential learning method that is based on Deep Q-network (DQN) and adjusted for fast convergence of multi-input problems. The results show better performance compared with the rule-based controller.