These slides were presented at the NAFEMS World Congress 2025, held in Salzburg, Austria from May 19–22, 2025.

Abstract

The interest in fuel cell vehicles has been recently increasing as they offer an alternative to battery electric vehicles for reducing emissions, in particular in cases where the boundary conditions for battery electric vehicles (availability and capacity for charging) are not met. Fuel cell vehicles rely on a complex energy management system to efficiently utilize the power generated by the fuel cell stack, manage the battery state-of-charge, and optimize overall vehicle performance. Designing an optimal energy management strategy for a fuel cell vehicle is however a challenging task due to the highly nonlinear and coupled dynamics involved, as the efficiency of a fuel cell depends not only on the amount of power which is requested, but also on the consistency of this demand. Reinforcement learning is a powerful machine learning technique that has shown great promise in solving complex control and decision-making problems. Unlike traditional control approaches that rely on detailed mathematical models, reinforcement learning allows an agent to learn an optimal control policy by interacting with the system and receiving feedback in the form of rewards or penalties. This paper investigates the application of reinforcement learning to determine an efficient and robust energy management strategy for a fuel cell vehicle. The goal is to maximize the vehicle's overall efficiency and driving range while maintaining the battery state-of-charge within desired limits. A reinforcement learning agent is trained to control the power delivered by the fuel cell to the battery, based on the current vehicle operating conditions such as power demand, battery state-of-charge, and fuel cell efficiency. By comparing the reinforcement learning approach against a traditional rule-based control strategy, this paper demonstrates that the RL agent is able to decrease the hydrogen consumption with 4% while the process to develop a reinforcement learning-based controller also requires much less manual effort. The results showcase the potential of reinforcement learning to enable adaptive and robust energy management for fuel cell vehicles. The reinforcement learning agent is able to learn a control policy that adapts to changing driving conditions and efficiently coordinates the fuel cell and battery subsystems. This leads to significant improvements in overall vehicle efficiency and driving range compared to the rule-based approach. Furthermore, the nature of reinforcement learning allows the energy management strategy to be quickly updated and deployed on different fuel cell vehicle platforms, reducing the engineering effort required.