Research On ROM Modeling Method Of NPP SBO Accident Based On Singular Value Decomposition

The safety issues of nuclear power plants have been widely concerned.The RISMC analysis method coupling deterministic safety analysis theory and probability safety analysis theory determines the key uncertainty factors affecting the accident process,extracts different accident scenarios,and uses a large number of simulation results to obtain statistical failure probability,so as to obtain the probabilistic safety margin of the nuclear power plant.This method can quantify the uncertainty of key parameters in accident scenarios,conduct dynamic safety analysis of nuclear power plants,and provide more cost-effective decisionmaking information for nuclear power plants.The RISMC method involves a multi-physical,multi-scale(temporal,spatial)problem.The overall computational load and cost of general RISMC analysis are very significant.To solve this problem,researchers have proposed the idea of using alternative models to replace complex thermal hydraulic simulation code,which can effectively reduce computational load and improve computational efficiency.Based on the idea of dimensionality reduction,this article adopts an alternative model construction method which is based on singular value decomposition and applies it to RISMC analysis.The specific process is: Latin hypercube sampling is selected as the forward sampling method to cover the input space scientifically.Taking the SBO accident of Daya Bay Nuclear Power Plant as the research object,determine the important influencing parameters and their distribution types of the power plant,select hazardous sequences for sampling calculation;Based on the calculation results of the thermal hydraulic code,typical time trends extracted from the results are used as orthogonal basis for reconstructing the time series using singular value decomposition method,and the expansion coefficients are obtained through orthogonal basis expansion.This thesis uses BP neural network to establish a function between input data and expansion coefficients,thus completing the establishment of alternative models and conducting case testing analysis on the models.This thesis uses the SBO accident of Daya Bay Nuclear Power Plant as the case study object combines the singular value decomposition method developed in this thesis to obtain approximate simulation results with faster computational efficiency.Compared with the IDW method in RAVEN software,the calculation results verify the effectiveness of the alternative model and achieve the goal of reducing calculation costs.The model can quickly reconstruct accident sequences using samples obtained from thermal hydraulic code,decreasing the calculation cost from 14 hours to 32 minutes for SBO accident analysis.