Fault detection for time-varying process in nuclear power plants based on improved moving window and dynamic kernel principal component analysis

In order to accurately identify the abnormal state of nuclear power plants and overcome the impact of its time-varying characteristics on the detection model, this study proposes a fault detection method based on k-means clustering improved moving window (MW) and dynamic kernel principal component analysis (DKPCA). Firstly, the system variables are grouped based on correlation analysis to effectively improve the sensitivity of the detection model to faults. Then, the corresponding DKPCA fault detection model is established for each group of variables, and local outlier factor and moving average filtering are used to eliminate false alarms of the model. In order to adapt to the time-varying characteristics of the system, a model update judgment mechanism based on K-means clustering and MW is introduced, thereby eliminating unnecessary updates of the model. The proposed method is applied to the real operation data of reactor coolant pump in a nuclear power plant, and different indicators are used to conduct the evaluation of the method. The test results indicate that the proposed improved DKPCA method achieves the average fault detection rate and false alarm rate of 88.80 % and 1.38 % across three abnormal signals while accurately identifying abnormal variables. Additionally, for time-varying signals, the proposed improved MW-DKPCA method achieves the average fault detection rate and false alarm rate of 85.89 % and 4.73 % across four long-term fault detection tasks, which can effectively eliminate unnecessary model updates while ensuring fault detection performance.