基于参数自回归算法的核电厂关键设备早期预警方法研究

本研究设计并开发了一种基于参数自回归算法的用于核电厂关键设备早期预警的方法，该方法创新性地引入了基于多维度时序数据的参数自回归算法，对设备正常运行状态下的参数进行估计，并通过与实测值分析比较来提取残差特征，从而实现了基于动态阈值的设备状态监测机制。此外，结合设备机理，本研究提出并采用了测点重要度的关键概念，通过对设备核心部件拆分建模，实现了对设备运行状态的识别、异常征兆的早期预警、故障部件的确定和相关报警事件的生成。本研究在AP1000核电机组的核心设备——反应堆冷却剂泵（简称主泵）上对设计开发的方法进行了测试验证，通过对主泵实际运行数据和异常事件的相关数据分析，结果表明，与现有设备状态监测方法相比，新建立的关键设备早期预警方法可以在早期发现相关设备的异常征兆，进行预警，并给出关键信息协助工程师进行故障的分析和定位，从而显著缩短了故障诊断和排故的时间，同时极大降低了关键设备监测的人力投入。 

Parameter Autoregression Algorithm-Based Early Warning Method for Critical Equipment in Nuclear Power Plants

A methodology based on the parameter autoregression algorithm is designed and developed for the early warning of critical equipment in nuclear power plants. The method innovatively introduces a parameter autoregression algorithm based on multi-dimensional time sequence data, which estimates the parameters under normal operation of the equipment and extracts the residual characteristics by comparing them with the measured values, thus realizing a dynamic threshold-based equipment condition monitoring mechanism. In addition, combined with the equipment mechanism, this study proposes and adopts the key concept of measurement point importance, and through the modeling of the core components of the equipment, the identification of the equipment operating state, the early warning of abnormal signs, the identification of faulty components and the generation of relevant alarm events are achieved. This study tests and validates the designed and developed method on the reactor coolant pump (hereinafter referred to as the main pump), the core equipment of AP1000 nuclear power unit. Through the analysis of the actual operation data and abnormal events of the main pump, compared with the existing equipment condition monitoring methods, the newly established early warning method for critical equipment can detect abnormal signs of relevant equipment at an early stage, produce early warning, and provide key information to assist engineers in fault analysis and localization, thus significantly shortening the time for fault diagnosis and troubleshooting, and greatly reducing the manpower input for critical equipment monitoring.