Abnormal Event Identification in Nuclear Power Plants Using a Neural Network and Knowledge Processing

Abstract

The combination of a neural network and knowledge processing have been used to identify abnormal events that cause a reactor to scram in a nuclear power plant. The neural network recognizes the abnormal event from the change pattern of analog data for state variables, and this result is confirmed from digital data using a knowledge base of plant status when each event occurs. The event identification method is tested using test data based on simulated results of a transient analysis program for boiling water reactors. It is confirmed that a neural network can identify an event in which it has been trained even when the plant conditions, such as fuel burnup, differ from those used in the training and when the analog data contain white noise. The network does not mistakenly identify the nontrained event as a trained one. The method is feasible for event identification, and knowledge processing improves the reliability of the identification.