Minor breaks fault detection in Nuclear Power Plants based on KPCA residual subspace

Abstract

Nuclear Power Plants (NPPs) are permitted a specific level of leakage during regular operating conditions for process reasons. This paper studies the application of residual subspace kernel principal component analysis and Kullback-Leibler Divergence (RSKPCA-KLD) in the fault detecting of minor breaks, addressing the current limitations of detection thresholds for such occurrences. First of all, given the traditional kernel principal component analysis (KPCA) ignores training data redundancy, preprocessing is implemented to eliminate redundant variables and decrease the training data volume, which contains Reduced KPCA, Analysis of Variance (ANOVA), and Pearson's correlation coefficient. Second, one probability-related nonlinear statistical monitoring model is constructed by integrating KPCA residual subspace with Kullback-Leibler Divergence (KLD), which measures the probability distribution changes caused by minor shifts. Third, considering the model's performance, the grid search is implemented to optimize hyperparameters, while a sliding window approach achieves local feature extraction. The experimental findings indicate that the equivalent diameters of detectable minor breaks have decreased by an order of magnitude relative to prior research, which improves the economics of NPPs.