核电多回路系统多源传感器异常检测的AAKR-SPRT方法  

Application of AAKR-SPRT to Sensor Anomaly Detection in Nuclear Power Multi-loop System

作　　者：谢述帅 成玮[1] 张乐[1] 聂泽琳 陈雪峰[1] 李芸[1] XIE Shushuai;CHENG Wei;ZHANG Le;NIE Zelin;CHEN Xuefeng;LI Yun(State Key Laboratory of Aerospace Power System and Plasma Technology,Xi'an Jiaotong University Xi'an,710049,China)

机构地区：[1]西安交通大学航空动力系统与等离子体技术全国重点实验室,西安710049

出　　处：《振动.测试与诊断》2025年第2期233-239,407,408,共9页Journal of Vibration,Measurement & Diagnosis

基　　金：国家重点研发计划资助项目(2019YFB1705403);国家自然科学基金资助项目(52105121);王宽诚教育基金会资助项目。

摘　　要：针对核电多回路耦合系统在升功率运行中异常传感器检测困难、检测延时及检测精度低等问题,提出了一种自联想核回归模型(auto-associative kernel regression,简称AAKR)与修正序贯概率比检验(sequential probability ratio test,简称SPRT)相结合的方法。首先,利用小波软阈值降噪方法对监测数据预处理,获取高质量的多源传感器解调信号;其次,采用AAKR构造传感器正常运行数据的估计值,并获取多源传感器测量值与估计值之间的残差;然后,运用滑动时间窗获取不同阶段残差向量的均值和方差,设计一种SPRT检测规则对传感器残差进行异常检测;最后,用核电一、二回路耦合系统模拟机实验数据进行方法验证与性能分析。结果表明,所提传感器异常检测方法的准确率达到99.52%,异常检测延时降低了81.73%,可有效提高现有核电厂传感器异常检测的稳定性。Aiming at the difficulties in abnormal sensor detection,detection delay and low detection accuracy in the power-up operation of nuclear power multi-loop coupling system,a method of combining self-associative kernel regression model auto-associative kernel regression(AAKR) with modified sequential probability ratio test(SPRT) is proposed for anomaly detection of multi-source sensors.First of all,the wavelet soft threshold denoising method is used to preprocess the monitoring data to obtain high quality demodulated signals of multisource sensors.Secondly,the estimated value of the normal operation data of the sensor is constructed by using the AAKR model,and the residuals between the measured value and the estimated value of the multi-source sensor are obtained.Then,the mean and variance of residual vectors in different stages are obtained by using sliding time window,and a new SPRT detection rule is designed to detect sensor residuals.Finally,the experimental data of nuclear power primary and secondary loop coupling system simulator are used for method verification and performance analysis.The results show that the accuracy of the proposed sensor anomaly detection method reaches 99.52%,and the anomaly detection delay is reduced by 81.73%,which can effectively improve the stability of abnormal detection of existing nuclear power plant sensors.

关 键 词：核电系统 传感器异常检测 自联想核回归 序贯概率比检验 小波阈值降噪 

分 类 号：TL362.3[核科学技术—核技术及应用] TH17[机械工程—机械制造及自动化]