Ecological Driving Strategy for Fuel Cell Hybrid Electric Vehicle Based on Continuous Deep Reinforcement Learning
Weiqi Chen, Guodong Yin, Yi Fan, Weichao Zhuang, Hailong Zhang, Jiankun Peng
Abstract
Fuel cell hybrid electric vehicles are essential approaches to achieve energy saving and emission reduction in transportation sector. The core idea of ecological driving is to reduce fuel consumption as much as possible on the premise of satisfying travel needs, which is a complex nonlinear multi-objective coupled optimization problem. Conventional algorithms have shortcomings such as poor optimization effect, heavy computational burden, and difficulty in online application. A novel integrated framework of ecological driving based on continuous deep reinforcement learning algorithm is proposed to bridge research gaps in this paper. Reward functions are designed to guide agent to optimize car-following performance, fuel consumption and change rate of output power of fuel cell system synchronously. In order to explore performance of the proposed method, optimization results with different weight coefficient values are compared and analyzed. Simulations under different driving cycles manifest excellent adaptability of the proposed method.