A reinforcement learning-based energy management strategy for fuel cell electric vehicle considering coupled-energy sources degradations

An effective energy management strategy (EMS) is crucial for fuel cell electric vehicles (FCEVs) to optimize fuel consumption and mitigate fuel cell (FC) aging by efficiently distributing power from multiple energy sources during vehicle operation. The Proton Exchange Membrane Fuel Cell (PEMFC) is a preferred main power source for fuel cell vehicles due to its high power density, near-zero emissions, and low corrosivity. However, it is expensive, and its lifespan is significantly affected by rapid power fluctuations. To address this issue, the proposed method of minimizing instantaneous cost (MIC) reduces the frequency of abrupt changes in the FC load. Additionally, by analyzing driving condition characteristics, the Ensemble Bagging Tree (EBT) facilitates realtime recognition (WCI) of composite conditions, thereby enhancing the EMS's adaptability to various operating conditions. This paper introduces an advanced EMS based on double-delay deep deterministic policy gradient (TD3) deep reinforcement learning, which considers energy degradation, economic efficiency, and driving conditions. Training results indicate that the TD3-based policy, when integrated with WCI and MIC, not only achieves a 32.6 % reduction in FC system degradation but also lowers overall operational costs and significantly accelerates algorithm convergence.