Real‐Time Energy Management Strategy for Fuel Cell/Battery Plug‐In Hybrid Electric Buses Based on Deep Reinforcement Learning and State of Charge Descent Curve Trajectory Control

In order to reduce the energy consumption of fuel cell/battery plug‐in hybrid electric buses and prolong the service life of fuel cell and power battery, this article proposes a multiobjective real‐time energy management strategy (EMS) based on deep reinforcement learning and state of charge (SOC) descent curve trajectory control. First, the demand curve based on power is derived from the operational data collected from a bus in Dalian, and a set of SOC reference decline curves based on mileage is formulated. Second, a multiobjective cost function is constructed to consider hydrogen consumption cost, power consumption cost, fuel cell lifespan, and power battery lifespan. Finally, a comprehensive dynamic decision Q network (CDDQN) framework based on double deep Q network (DDQN) is established, a series of deep reinforcement learning EMS that integrate CDDQN with SOC trajectory control are designed, and they are validated through experimental analysis. The results demonstrate that this strategy exhibits excellent real‐time performance and economic efficiency. Compared with the comparison algorithms rule algorithm, equivalent consumption minimization strategy, finite‐step dynamic programming (DP), and DDQN proposed herein, the economy is increased by 15.08, 13.48, 8.81, and 3.07%, respectively, and reaches 94.00% of the economy of the ideal optimal solution DP.