Pipe network leakage anomaly identification method and system with constraint semi-supervised clustering

Abstract

This invention relates to the field of water supply network anomaly identification technology, specifically to a constrained semi-supervised clustering method and system for identifying water leakage anomalies in water supply networks. The method includes: constructing a convex loss function that fuses sets of constraint pairs that must be linked, sets of constraint pairs that cannot be linked, and a regularization loss term; solving for the optimal feature weight vector; calculating the fused Mahalanobis distance to construct a fused distance matrix; performing clustering iterations; in each iteration's cluster assignment, grouping similar samples into the same cluster and dissimilar samples into different clusters until convergence, obtaining the final clustering result; calibrating the operating condition category of each cluster; and outputting the identification results for normal operating conditions, water leakage anomalies, or unknown anomalies. This method improves the accuracy, recall, and clustering stability of water leakage anomaly identification through the synergistic effect of business constraint guidance, pre-optimized convex feature weights, and statistical distance correction, while suppressing interference from strong noise and feature overlap, achieving a deep fusion of business priors and data-driven approaches.

Images