Calibration of roughness coefficient for long-distance water supply systems with multi-branch pipelines.

This article focuses on a novel calibration framework for hydraulic models of long-distance water supply systems, addressing the common ill-posed problem caused by limited measurement data. The proposed method integrates artificial neural networks (ANN) to estimate flow rates and pressures at unmeasured nodes with adaptive particle swarm optimization (APSO) to optimize pipe roughness coefficients, incorporating physical law-based regularization into the ANN loss function. Tested on a large-scale Chinese water supply system, the framework demonstrated improved calibration accuracy and robustness against measurement uncertainties compared to traditional methods, with sensor placement and regularization shown to significantly influence performance. The study highlights the potential of combining ANN and APSO for effective hydraulic model calibration under data scarcity and suggests future work on incorporating prior knowledge and advanced algorithms to enhance reliability and applicability.