With the rapid advancement of urbanization, the operating conditions of rail transit vehicles have become increasingly complex and stochastic. Consequently, conventional energy management strategies based on fixed control rules often fail to achieve optimal performance. This study proposes an adaptive energy management strategy for hydrogen-powered hybrid systems in rail transit, informed by real-time recognition of operating condition features. The proposed method classifies the diverse load conditions into three representative modes—low-speed braking, economic cruising, and high-speed operation—based on distinctive dynamic characteristics. A hierarchical and coupled global sensitivity analysis framework, incorporating an online particle swarm optimization algorithm, is developed to determine the optimal multi-objective weighting under varying conditions. This enables real-time adaptive multi-objective energy management tailored to dynamic scenarios. Comparative results demonstrate that, relative to fixed weight strategies, the proposed method significantly reduces fluctuations in the state of charge (SOC) of the lithium battery while maintaining comparable fuel cell efficiency and hydrogen consumption. Moreover, all three key performance metrics exhibit substantial improvements compared to conventional ECMS.