［Purposes］How to rapidly and effectively dispatch rescue forces in the initial stage of an incident, while ensuring dynamic and orderly coordination in subsequent operations, remains a core challenge in large-scale maritime life-saving. Considering rescue helicopters, vessels, and coast guard ships, this study addresses scenarios characterized by large numbers of distressed persons, limited operational space, and complex marine environments, and proposes a two-stage rescue force scheduling optimization model based on optimization strategies and reinforcement learning.［Methods］In the first stage, to quickly obtain on-site information and evacuate critically endangered individuals, a bi-objective scheduling model is developed that balances maximizing the cumulative comprehensive score of rescue forces and minimizing the number of deployed units, enabling rapid and rational initial dispatch. To achieve dynamic and orderly rescue for the remaining distressed persons, the second stage incorporates on-site capacity constraints and an environmental risk assessment mechanism, constructing a dynamic scheduling model solved using a reinforcement learning-based algorithm to optimize subsequent resource allocation.［Findings］A large cruise ship accident in the Taiwan Strait is used as a case study. Simulation results show that the proposed two-stage scheduling approach aligns well with real operational conditions, achieving rapid initial response and dynamic, orderly follow-up rescue. ［Conclusions］The findings offer practical support for efficient and coordinated deployment of rescue resources in large-scale maritime life-saving operations.