Research on PIRT Automatic Generation Method based on Deep Learning Algorithm

The Best Estimate Plus Uncertainty (BEPU) analysis method is one of the main approaches for nuclear safety review. However, conducting sensitivity and uncertainty analyses with best-estimate code requires high calculation costs. Artificial intelligence-based surrogate models can significantly improve analytical efficiency. In this paper, the steam generator tube rupture (SGTR) accident in a nuclear power plant is used as a case study. A surrogate model is established using a deep neural network algorithm, coupled with the sensitivity and uncertainty analysis code DAKOTA. Based on accident acceptance criteria, accident input and output variables are identified, and Sobol sensitivity indicator is used for sensitivity analysis. The analysis results provide an importance ranking of input parameters based on sensitivity indicator, and a secondary-ranked Phenomena Identification Ranhing Table (PIRT) is generated. The study demonstrates that the DNN-based surrogate model can accurately predict the variation trends of critical safety parameters, can be used for sensitivity analysis of these parameters to obtain importance ranking of parameters, and can automatically generate a PIRT for importance ranking of parameters.