As a high-performance energy platform, ocean nuclear power plant operates in complex and dynamic environments, requiring optimization of operational parameters to balance safety and maneuverability. In order to meet the demands of multi-criterion optimization, achieve precise adjustment of operational parameters, and enhance the overall performance of the system, an evolutionary multi-criterion optimization framework based on surrogate models is proposed in this study. Optimization objectives are defined through requirement analysis, and an evaluation index system centered on safety and maneuverability is established. A surrogate model of the system was constructed using deep learning combined with simulation operation data, enabling accurate virtualization and mapping of the physical system, for efficient operational state prediction under complex conditions. An evolutionary multi-criterion optimization framework is developed, and a dynamic optimization strategy based on system characteristics is constructed. Results show that the integration of the surrogate model and evolutionary multi-criterion optimization algorithm significantly improves the computational efficiency and optimization accuracy, and provides reliable technical support for the operation of ocean nuclear power plants. The proposed optimization method plays a vital role in enhancing the comprehensive performance of ocean nuclear power plants, while also providing an innovative approach for the optimization design of similar complex systems, with broad potential for engineering applications.