Comparative Study of Cluster Analysis Methods in Dynamic Safety Analysis of Nuclear Power Plants

Abstract In nuclear power plant safety analysis, employing dynamic safety analysis methods such as Dynamic Event Trees (DET) to analyze individual initiating events often generates numerous scenarios, posing challenges in extracting useful information. This study utilizes clustering analysis methods to categorize these accident scenarios and extract their main features, thereby simplifying the analysis process and enhancing efficiency. Different clustering analysis methods can yield varying results, affecting the ability to reflect accident processes and extract key features. To ensure that clustering analysis better reflects accident processes and extracts the main features of accidents, this study compares and investigates two clustering analysis methods: one that considers only extreme parameter values of accidents and another that considers parameters throughout the entire accident process. In the method focusing solely on extreme parameter values, key parameters are extracted as clustering analysis parameters from each accident scenario. In contrast, the method considering parameters throughout the entire accident process treats accident scenario parameters as time-series data and applies Dynamic Time Warping (DTW) algorithm for time-series clustering. Using the Steam Generator Tube Rupture (SGTR) accident as an example, this study applies both methods to analyze SGTR accidents and obtains corresponding results. The characteristics and differences between the two methods are analyzed.