基于孪生模型的堆芯自给能中子探测器信号异常检测

Anomaly Detection of Core Self-Powered Neutron Detector Based on Twin Model

堆芯自给能中子探测器(SPND)的健康状态直接影响反应堆安全运行。充分考虑核反应堆内不同位置SPND间的量测关联性,提出一种基于孪生模型的堆芯SPND信号异常检测方法,利用SPND运行状态历史数据,通过随机森林回归算法(RFR)提取邻域SPND的量测信号特征,为SPND构建与其物理实体一致输出的孪生模型。孪生模型与探测器实体共生运行,通过计算探测器观测值与孪生模型估计值之间的残差,作为SPND信号异常检测判据,以此实现对单点和多点异常SPND的辨识和定位。实验表明,本文所提孪生模型的预测误差在1×10^(-11)量级,具备极高的输出一致性。针对多种不同SPND信号异常状态,能够获得99%以上的辨识精度,准确地判别单点及多点异常SPND,对于提高堆芯中子注量率测量系统状态监测的可靠性和安全性具有较高的参考价值。

Self-powered neutron detector(SPND)is an important nuclear sensing device in the core,whose health status affects the safe operation of the reactor directly.Considering the measurement correlation between SPNDs at different positions in the reactor,a twin model-based anomaly detection method for SPND signals in the core was proposed in this paper.The characteristics of measurement signals of neighboring SPNDs were extracted by the Random Forest Regression(RFR)algorithm based on the historical operation data of SPND,and a twin model was built for SPNDs which outputs the same as its physical entity.Twin model and entity sensing coexisted.The residual error between the actual observation value of SPND and the twin model estimation value was calculated to serve as the anomaly detection criterion,which realized the identification and location of single-point and multi-point SPND anomalies.The experiments show that the prediction error of the twin model proposed in this paper attains the order of 1×10^(−10),which has a very high output consistency.The identification accuracy of anomaly detection can reach over 99%under various abnormal states of SPND signals,and the single-point and multi-point abnormal SPND can be accurately identified,which has a high reference value for improving the reliability and safety of the state monitoring of neutron flux measurement system in the core.