Severe snowstorms pose multiple threats to high-speed rail systems, including sudden drops in track friction coefficients, icing of overhead contact lines, and reduced visibility. These conditions can trigger dynamic risks such as train speed restrictions, cascading delays, and operational disruptions. Addressing the limitations of traditional scheduling methods in spatio-temporal modeling during blizzards, real-time multi-objective trade-offs, and high-dimensional constraint solving efficiency, this paper proposes a collaborative optimization approach integrating temporal forecasting with deep reinforcement learning. A dual-module LSTM-PPO model is constructed using LSTM (Long Short-Term Memory) and PPO (Proximal Policy Optimization) algorithms, coupled with a composite reward function. This design collaboratively optimizes punctuality and scheduling stability, enabling efficient schedule adjustments. To validate the proposed method’s effectiveness, a simulation environment based on the Lanzhou-Xinjiang High-Speed Railway line was constructed. Experiments employing a three-stage blizzard evolution mechanism demonstrated that this approach effectively achieves a dynamic equilibrium among safety, punctuality, and scheduling stability during severe snowstorms. This provides crucial decision support for intelligent scheduling of high-speed rail systems under extreme weather conditions.