Answers are grounded in this paper and cite their sources.
This study presents a highly sensitive methodology for the early detection of incipient leaks in Water Distribution Networks (WDNs), addressing the critical operational challenge where minute anomalies are obscured by severe background noise and complex, non-polynomial cyclical trends. To overcome the limitations of conventional filtering, we introduce a tailored Empirical Wavelet Transform (EWT) framework specifically engineered to suppress these nonlinear baseline dynamics while preserving weak anomaly signatures. Following this preprocessing, a novel dual-criteria detection scheme evaluates the detrended signals. This mechanism uniquely combines Kullback-Leibler (KL) entropies divergence, which captures subtle probabilistic shifts in signal distributions, with Root Mean Squared Error (RMSE) evaluation, which quantifies reconstruction deviations. The synergistic integration of these metrics enables highly reliable discrimination between normal operational variations and actual leak-induced anomalies. The efficacy of the proposed framework was rigorously validated across a comprehensive 500-scenario benchmark dataset based on the Hanoi WDN. The method successfully identified extremely small, developing leaks, achieving an exceptional F1-score of 96.17% and a sensitivity of 92.63% in distinguishing between leaky and leak-free scenarios. Crucially, the system maintained a 0.00% false alarm rate, significantly outperforming contemporary state-of-the-art deep learning and statistical baselines. By effectively eliminating false positives and capturing early-stage infrastructure failures, this automated detector provides precise, actionable decision support. The proposed methodology directly facilitates intervention prioritization, minimizes operational cost-risk trade-offs, and contributes significantly to sustainable water resource management and the reduction of urban water losses.
No takes yet. Share an insight, caveat, or question.
Damonti et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: