Hybrid risk evaluation and prediction method for nuclear power plant using RISMC and transformer-based surrogate model

The safe operation of nuclear power plants imposes strict requirements on the accurate prediction of key parameters and risk assessment, particularly under complex operating conditions and accident scenarios, where the dynamic evolution of key parameters directly impacts operators’ situational awareness and decision-making. This study proposes a hybrid risk evaluation and prediction method based on Risk-informed Safety Margin Characterization(RISMC) methodology, integrating high-fidelity thermal–hydraulic simulation with a Transformer-based surrogate model, significantly expanding the application of RISMC to the operation scenario of nuclear power plants. By combining time-series prediction with regression models, the proposed method enables high-accuracy prediction of the dynamic evolution of critical parameters, such as steam generator water level and pressure, while dynamically quantifying the impact of various intervention strategies on reactor shutdown risk. This approach effectively addresses the challenges of existing risk assessment methods in computational efficiency and performance during plant operation. A Partial Loss-of-Feedwater (PLOFW) scenario is used as a case study to verify the method’s effectiveness in dynamic prediction, risk quantification, and intervention strategy evaluation. The results demonstrate that the proposed method significantly outperforms prediction accuracy and computational efficiency. By precisely quantifying dynamic safety margin, the method significantly enhances operators’ situational awareness and decision-making capabilities during abnormal events and incidents, providing a basis for optimizing intervention strategies. This approach not only ensures the safe and flexible operation of nuclear power plants but also balances safety and economic efficiency.