Reinforcement Learning-Based Adaptive Hierarchical Equivalent Consumption Minimization Strategy for Fuel Cell Hybrid Engineering Vehicles

Abstract

To enhance the operational efficiency of fuel cell engineering vehicles in transportation, reliable energy management strategies (EMSs) are essential for optimizing fuel consumption and power distribution. In this paper, we propose a novel energy management framework that utilizes a reinforcement learning-based adaptive hierarchical equivalent consumption minimization strategy (ECMS) to regulate fuel cell/battery hybrid system. The structure integrates deep Q-network (DQN), fuzzy logic, and ECMS algorithms and employs a long short-term memory neural network for working condition prediction. By combining DQN with the equivalence factor obtained using the battery state of charge penalty function and adjusting it using a fuzzy logic controller, the stability of the subsequent ECMS is enhanced. In a simulation environment, the proposed EMS achieves a 97.44% fuel economy compared to the dynamic programming-based global optimized EMS. Experimental findings indicate that the hierarchical ECMS effectively decreases the equivalent hydrogen consumption by 3.38%, 9.12%, and 16.39% compared to the adaptive ECMS, DQN-based ECMS, and classic ECMS, respectively. Therefore, the proposed methodology offers superior economic benefits.