인공지능을 활용한 비정상 상황에서 잔여 트립 시간 예측 시스템 개발

Abstract

When an abnormal state occurs in the Nuclear Power Plants (NPPs), operators perform mitigation actions to prevent the reactor trip and restore normal conditions. However, in abnormal states, the operator’s tasks progress urgently checking many monitoring variables within a limited time. Consequently, operators may feel psychological pressure, which can lead to human errors. If mitigation actions fail due to human errors by the operator, the status of the NPPs may worsen and a reactor trip may occur. Furthermore, sudden shutdown by the reactor trip can cause the failure of numerous facilities in the NPPs. In addition, once the NPPs have been stopped, it takes a long time to restart, potentially leading to significant economic losses. Accordingly, in this thesis, a remaining trip time prediction system was developed as part of an operator support system with the goal of assisting operator decision-making in abnormal states, aiming to prevent potential human errors. The remaining trip time prediction system is implemented through an algorithm that uses artificial intelligence (AI) and explainable AI (XAI) methods such as autoencoder, light gradient boosting machine, and shapley additive explanation. The remaining trip time prediction algorithm provides diagnosis information about the abnormal state that has occurred and prediction information about the remaining time until the reactor trip. In addition, the XAI was used to provide the rationales for the remaining trip time prediction result and to improve the reliability of the An interface was developed to efficiently provide the results of the remaining trip time prediction algorithm. The interface can provide results of the remaining trip time prediction algorithm in real time regarding the current status of the NPPs. Therefore, the interface indicates the change in the results of the remaining trip time prediction algorithm according to the operator's mitigation actions. As a result, the remaining trip time prediction system provides diagnosis and prediction information and is expected to be helpful to the operator's tasks in abnormal states.