基于时变多块主元分析法的核电系统故障诊断

In nuclear power systems, usually, processes are complex and time dynamic, and a large number of measured and manipulated variables are highly correlated. As a dimension reduction technique for capturing strong correlation underlying in the process measurements, principal component analysis (PCA) is widely applied. However, as a glocal and static modeling method, PCA based fault detection technique often fails to extract local system characteristics and also has the model mismatch problem for dynamic system diagnosis. In this case, this work proposes an time varying PCA method purely based on operation data. The proposed method constructs time varying PCA models according to different changing operation conditions. The process variable space is partitioned into several sub-feature spaces, and the constructed sub-block models can not only reflect the local behavior of the process change but also contribute to enhanced fault detection through the combination of local fault detection results. The proposed method is demonstrated in a nuclear power system, and the results show that the novel method is much better in fault detection and isolation comparing to conventional PCA method.