Driving-Behavior-Aware Optimal Energy Management Strategy for Multi-Source Fuel Cell Hybrid Electric Vehicles Based on Adaptive Soft Deep-Reinforcement Learning
Haochen Sun, Fazhan Tao, Zhumu Fu, Aiyun Gao, Longyin Jiao
Abstract
The majority of existing energy management strategies (EMSs), merely considering external driving conditions, often allocate demand power in an irrational way, resulting in a waste of energy and a short service life of power sources. Therefore, it is necessary to integrate driving behavior in EMS to reduce the fuel consumption and improve the lifespan of power sources. In this paper, a driving-behavior-aware adaptive deep-reinforcement-learning (DRL) based EMS is proposed for a three-power-source fuel cell hybrid electric vehicle (FCHEV). To fully utilize each power source, a hierarchical power splitting method is adopted by an adaptive fuzzy filter. Then, a high-performance driving behavior recognizer is employed, and Pontryagin’s minimum principle (PMP) method is used to compute the optimal equivalent factor (EF) of each driving behavior. To realize a trade-off between global learning and real-time implementation, an improved multi-learning-space DRL-based algorithm, applying driving-behavior-aware adaptive equivalent consumption minimization strategy (A-ECMS) and soft learning mechanism, is proposed and verified by a series of simulations. Simulation results show that, compared with the benchmark method ECMS, the proposed P-DQL method can reduce the hydrogen consumption by 49.9% on average, and the total cost to use by 31.4%, showing a promising ability to increase fuel economy and reduce hydrogen consumption and the total cost to use of FCHEV.