On-site simulator-based initiating event detection and identification in NPPs

Abstract

Installing safety components to prevent a severe accident from developing from an initiating event is essential. Although various approaches have been presented to detect and identify initiating events, their development relies on the data simulated by computer codes like the Modular Accident Analysis Program. This study focuses on developing approaches to detect and identify initiating events using the data generated by a nuclear power plant (NPP) on-site simulator. By relying on historical data to construct an empirical model that can predict normal operation sensing readings, an abnormal event that leads to excessive discrepancies between the acquired and predicted readings can be detected. Several single- and multiple-sensor feature extraction methods have been presented to find low-dimensional representations for initiating events to facilitate accurate event identification. The results from detailed experiments containing data of 13 event categories with 185 events generated by the Maanshan NPP plant simulator demonstrated the feasibility of developing on-site simulator-based schemes for detecting and identifying initiating events.