Carbon-Aware Hydraulic Modeling: A Machine Learning Framework for Real-Time Pressure Correction in Urban Water Systems

Accurate pressure prediction water distribution systems is essential optimising pump operations, reducing energy use, and achieving carbon mitigation targets. This study evaluates the performance of conventional hydraulic models, Hazen Williams (HW) and Darcy Weisbach (DW), against multiple machine learning (ML) approaches, including Random Forest (RF), Gradient Boosting Regressor (GBR), and Multilayer Perceptron (MLP), using field measurements from a coastal urban network in Thailand. Model accuracy was assessed using the mean absolute error (MAE), root mean square error (RMSE), and the coefficient of determination (R²), along with estimates of CO₂ emissions associated with excess pumping. Results indicate that the HW baseline produced an MAE of ۲.۸۱ m, corresponding to ~۲.۴۵ tonnes/month of avoidable CO₂ emissions, while DW exhibited substantially lower accuracy (MAE = ۱۰.۷۲ m). The RF model achieved the best generalisable performance (MAE = ۰.۳۴ m, R² = ۰.۹۹۹۵) and reduced avoidable CO₂ emissions by over ۹۰% compared toHW overestimates pressures during peak demand periods, whereas RF maintained low errors across all nodes and hours. The findings demonstrate that integrating ML-based correction with hydraulic simulation , offering dual benefits in operational efficiency and climate mitigation. The proposed approach to other urban networks can water utilities meet net-zero carbon commitments

کلیدواژه ها:

Water Distribution Systems ، Hydraulic Modeling ، Machine Learning ، Carbon footprint reduction ، Urban infrastructure Urban infrastructure