VULNERABILITY ANALYSIS OF NPP EQUIPMENT BASED ON NEURAL NETWORK

The vulnerability analysis is a vital part of the seismic probabilistic risk assessment of nuclear power plants. However, due to the complexity of nuclear power structures and the larger calculation scale, the vulnerability analysis of NPP equipment is very time consuming when considering soil-structure interaction (SSI). In order to develop an efficient vulnerability analysis method, this paper adopts a partition calculation method applied to NPP SSI analysis, and establishes an artificial neural network (ANN) using limited SSI analysis results to substitute the FEM process. Based on the regression method with log-normal assumption and Monte Carlo method to analyze the equipment vulnerability. The ANN numerical simulation includes the following contents. (1) Establish the best ANN model through cross-validation to substitute the FEM process, and the most relevant ground motion parameters are selected as the ANN input based on the semi-partial correlation coefficient. (2) Quantification and investigation of the ANN prediction uncertainty. It includes the aleatory uncertainty caused by the simplification of the seismic inputs and the epistemic uncertainty from the limited size of the training data. (3) Computation of fragility curves with Monte Carlo method and the regression method with log-normal assumption based on the prediction data of ANN model. This paper explores the impact on fragility curves induced by different seismic intensity measures and uncertainty of soil material. Meanwhile, the results verify the basic rationality of the lognormal assumption and provide a possible direction for the vulnerability analysis of NPP equipment.