A Q-Learning-Based Energy Management Strategy for a Three-Wheel Multi-Stack Fuel Cell Hybrid Electric Vehicle

This paper addresses the design of an adaptive energy management strategy (EMS) for a multi-stack fuel cell hybrid electric vehicle (MFC-HEV) using reinforcement learning (RL). The proposed strategy has two operating layers. In the first layer, the models of the FC and battery are updated online by recursive least squares (RLS) and then the updated characteristics are used by Q-Learning algorithm for distributing the power among the three FCs and the battery pack. The performance of the suggested strategy is compared with dynamic programming (DP) under a real driving cycle. The results show that there is almost a 7% performance difference in terms of total cost (power sources' degradation and hydrogen consumption) between the proposed strategy and DP under the considered driving cycle.