A Comparative Study of Energy Management Strategies for Fuel Cell Hybrid Vehicles Based on Deep Reinforcement Learning

Abstract

The energy management strategy (EMS) is the top priority to ensure the safe and efficient operation of fuel cell hybrid vehicles. Nowadays, EMS based on deep reinforcement learning (DRL) has become a research hotspot. However, there is a lack of a unified comparison benchmark for DRL-based EMSs. Most DRL-based EMSs has not discussed the impact of algorithm hyperparameters, and has not provided a comprehensive evaluation of indicators including fuel cost, aging cost, and efficiency. In this paper, five EMSs based on different DRL methods are designed, and a multi-objective reward function that integrates equivalent hydrogen consumption, fuel cell degradation, battery state-of-charge fluctuation and its safe working range is designed. First, the hyperparameters of DRL-based EMSs are determined based on the convergence performance in the training process. The weight coefficients of the multi-objective reward function are determined based on the average equivalent hydrogen consumption. Then the comprehensive performance of the above-mentioned DRL-based EMSs are compared horizontally. Finally, six driving conditions are used as test sets to explore the adaptability of DRL-based EMSs. The results show that DRL-based EMSs can be effectively applied in real-time environments, and different DRL algorithms show different performance in applications, which can provide valid guidance for researchers to use DRL algorithms in EMS.

Keywords: Energy management strategy, Fuel cell hybrid vehicles, Fuel cell, Lithium battery, Deep reinforcement learning