Volume 39: Energy Transitions toward Carbon Neutrality: Part II

Energy-Saving and Thermal Comfort Control of Electric Vehicle Air Conditioning Systems with Deep Reinforcement Learning Shuai Dai, Kuining Li, Jiangyan Liu, Yi Xie

Abstract

This study employs deep reinforcement learning algorithms, including Deep Q-Network, Deep Deterministic Policy Gradient, Twin Delayed Deep Deterministic Policy Gradient, and Soft Actor-Critic, to control the air conditioning system of electric vehicles to improve thermal comfort and reduce energy consumption. Additionally, random adjustments to environmental temperature and solar radiation intensity during the training process are made to enhance the algorithms’ applicability. The results demonstrate that these algorithms significantly reduce energy consumption while maintaining thermal comfort. Notably, the Deep Deterministic Policy Gradient algorithm achieves an impressive 37.6% reduction in energy consumption. Comparative analysis among the algorithms reveals that Deep Q-Network, Deep Deterministic Policy Gradient, and Twin Delayed Deep Deterministic Policy Gradient exhibit relatively stable control behaviors. In contrast, the Soft Actor-Critic algorithm’s compressor control curve exhibits more significant fluctuations, potentially leading to mechanical wear. Deep Q-Network, Deep Deterministic Policy Gradient, and Twin Delayed Deep Deterministic Policy Gradient algorithms consistently demonstrate effective thermal comfort control and energy-saving performance in various operating conditions.

Keywords deep reinforcement learning, electric vehicle, air conditioning system, thermal comfort