基于数据驱动的核动力系统异常检测及分析方法研究

针对核动力系统在线异常检测存在故障样本稀少且不完备的现实问题，借鉴安全运行域的概念，基于逻辑距离计算的思路，提出一种基于正常运行数据驱动的核动力系统异常检测方法，并以某核动力系统常用运行工况历史数据为对象，对算法进行了试验验证。结果表明，设计算法能有效检测系统异常和故障，具有良好的可靠性和可解释性，并且检测力度具有可调节性。 

Study on Data Driven Anomaly Detection and Analysis Algorithm for Nuclear Power Systems

Aiming at the problem that there are few and incomplete fault samples in nuclear power system on-line anomaly detection, referring to the concept of Safe operation domain, and based on the idea of logical distance calculation, this paper proposes a nuclear power system anomaly detection algorithm based on normal operation data. Taking the historical data of common operating conditions of a nuclear power system as the object, numerical experiments were carried out to verify the algorithm. The results show that the design algorithm can effectively detect system anomalies and faults, with good reliability and interpretability, and the detection strength can be adjusted.