Wireless monitoring and machine learning-based prediction of structural variations in urban water supply networks

Abstract

The safe and stable operation of urban water supply networks is critical to ensure urban functionality and sustainable economic development. However, the frequent occurrence of pipeline damage incidents highlights the limitations of traditional monitoring methods based on water quality, flow rate, and pressure. This study employs a wireless monitoring system that overcomes the constraints of traditional monitoring approaches by collecting real-time multi-source data on pipeline structure and operating environments. Support vector regression, feedforward neural networks, and physics-informed machine learning model (PIML) were used to quantitatively analyze the impact of various environmental factors on pipeline structural deflection angles and to train a high-accuracy machine learning prediction model. The results reveal significant variations in earth pressure, soil structure, temperature at the pipe crown and invert, and pore water pressure during pipeline operation, reflecting characteristics of backfilling, foundation settlement, and groundwater dynamics. Machine learning models trained on the monitoring data exhibited outstanding predictive accuracy, with PIML achieving the highest performance—showing an R² of 0.985 and a 96.9% overlap between predicted and actual distributions. Furthermore, interpretative analyses identified soil structure variation and burial depth as the primary driving factors influencing pipeline structural deflection angles. Building on this, monitoring strategies can be optimized to provide robust support for improving the safety and operational efficiency of urban water supply pipeline systems.