Abstract: To address the challenges of equivalent factor regulation hysteresis and insufficient driving cycle adaptability in energy management strategies for fuel cell vehicles under dynamic operating conditions, the study develops a dual-layer predictive and driving cycle recognition-based optimal control framework. We first propose a forecast-adaptive equivalent consumption minimization strategy incorporating driving demand prediction. This approach establishes a neural network-based driving cycle prediction model and integrates time-domain rolling optimization to update equivalent factors, thereby improving powertrain robustness against stochastic disturbances. Building upon this foundation, the enhanced PF-AECMS strategy further incorporates a dual-layer prediction architecture and equivalent factor smoothing constraints, achieving a 7% reduction in fuel cell power fluctuation amplitude while limiting equivalent factor overshoot to ±8%. Simulation results demonstrate that the PF-AECMS strategy exhibits superior control performance compared with both F-AECMS and conventional AECMS. The proposed strategy achieves a 2.4 s recovery time for battery state of charge(SOC) to reach its target value, representing a 27.3% reduction compared with F-AECMS, with a terminal SOC deviation of merely 0.3% from the target. Real-vehicle experimental validation reveals that PF-AECMS improves economic performance by 4.3% and 27.6% compared to F-AECMS and rule-based strategies, respectively, confirming its optimization effectiveness under dynamic conditions. This research provides a novel methodology with significant engineering application value for fuel cell vehicle energy management in dynamic operating scenarios, offering both theoretical innovation and practical implementation potential.

Key words: fuel cell vehicles, adaptive energy management strategy, dual-layer prediction, driving condition recognition, equivalent factor optimization