Abstract: Power grid infrastructure in mountainous regions of China is generally underdeveloped, and the long-distance external power supply for electrified railways adversely affects their operational reliability and energy efficiency. To address this,this paper proposes a data-driven energy optimization dispatch strategy for railway power supply systems.First, aflexible railway power supply system topology based on energy routers is constructed, integrating photovoltaic (PV) and energy storage systems. Short-term power forecasting for PV generation and traction loads is performed using recurrent neural networks (RNNs), with typical scenarios obtained through scenario-based methods to characterize uncertainties. Subsequently, considering the energy balance and safety constraints of the flexible traction power supply system, an optimization model aiming to minimize operational costs is established. Optimal energy dispatch is achieved by coordinating the operation of PV systems, energy storage systems, and energy routers. Finally, case study simulations analyze the system's optimized dispatch outcomes, utilization rate of regenerative braking energy, and daily electricity costs, further validating the effectiveness of the proposed strategy and model.