Safety Evaluation and Optimization of Nuclear Power Plant Under Adaptive Sampling

Abstract At present, the safety analysis methods used worldwide in nuclear power plants (NPPs) are divided into Deterministic Safety Analysis (DSA) and Probabilistic Safety Analysis (PSA). The traditional DSA is based on a concept of Defense-in-depth, which lacks the correct grasp of the state of NPPs and the economic consideration. The static fault tree and event tree methods used in PSA analysis are difficult to simulate the dynamic timing sequence. Previous studies on the safety margin of NPPs have mostly focused on the evaluation methods for the perspective of regulation, but there are no consensus on how to support the risk-informed decision-making of nuclear power plant (NPP). In addition, traditional DSA often only considers the epistemic uncertainty and does not include the random uncertainty such as random failure of equipment. In this way, the estimated safety margin is not accurate enough. In order to meet the comprehensive improvement of safety and economy in the Generation IV reactor, this paper is based on the Risk-informed Safety Margin characterization (RISMC) methodology, and focuses on the issues such as the large amount of calculation and long calculation time due to the requirements of high-precision under this methodology. A surrogate model training method based on Support Vector Machine (SVM), K nearest neighbor algorithm (KNN) and adaptive sampling strategy is proposed. During the process, key uncertainty parameters are screened through sensitivity analysis and the parameter dimensions are reduced. The surrogate model was then applied to RISMC methodology and combined with the Best Estimate Plus Uncertainty (BEPU) analysis to calculate the safety margin of NPPs. Finally, a Small-break Loss of Cooling Water Accident (SLOCA) is used as a case to demonstrate this method. By comparing the computation time of the surrogate model combined with adaptive sampling with a high-fidelity thermal-hydraulic model, it was found that this method can significantly reduce computation time. This provides technical support for applying RISMC method to engineering operation of nuclear power plants.