SEAMLESS LEVEL 2/LEVEL 3 DYNAMIC PROBABILISTIC RISK ASSESSMENT CLUSTERING

摘要

This paper discusses the work which has been conducted for a seamless Level 2 to Level 3 dynamic probabilistic risk assessment (PRA) station blackout scenario for a series of MELCOR input decks for a 3-loop pressurized water reactor with a sub-atmospheric dry containment. This work is an extension of past dynamic PRA experiments and includes a Level 3 MELCOR Accident Consequence Code Systems, Version 2 (MACCS2) analysis. Various parameters within the MELCOR analysis have distributions (e.g. creep rupture) which can significantly affect the overall accident progression and environmental release. To further reduce conservatism, an updated containment fragility curve was incorporated. Since the amount of data produced from a dynamic PRA may be difficult to analyze using current Level 2 binning processes, the concept of 'data mining' provides a methodology to extract useful information. A post-processing tool has been developed by the Ohio State University to data mine using a mean-shift methodology to cluster scenarios. The scenarios are aggregated according to information on system components (e.g., valves fail to close) and system process variables such as pressure and temperature in the reactor coolant system. Clustering of the results assists the analyst in identifying those scenarios that have the greatest contribution to risk.