설명 가능한 인공지능을 활용한 원전 통합 비정상 진단 시스템 개발

Abstract

In the Nuclear Power Plants(NPPs), keeping a normal state during operation is very important for safety and economics. If NPPs enters an abnormal state, the operators must take mitigation measures to bring it back to a normal state. Prior to taking mitigation measures, the operators must conduct an incident diagnosis of the current condition. However, in diagnosing an incident, the operators feels stress and psychological burden, which may lead to human error. If a false diagnosis is made due to human error, the NPPs can go out of control. In this thesis, an integrated abnormal diagnosis system in NPPs was developed to perform the reduction of operators’ human errors. For the system development, data collection using Compact Nuclear Simulator, data pre-processing, integrated abnormal diagnosis algorithm, and explanation interface development were performed. The system is implemented by applying Artificial Intelligence(AI), eXplainable AI(XAI), and rule-based system, and provides the results of applied methodologies to operators through an interface. By integrating these series of processes, an integrated abnormal diagnosis system in NPPs was developed. Providing diagnosis results through AI is expected to reduce human error, and diagnosis basis interpreted through XAI is expected to increase the reliability of artificial intelligence. In addition, it is expected that efficient information delivery will be possible by providing the diagnosis result through an intuitively configured explanation interface.