DefinePK

Abstract

Urban water distribution systems often suffer from inefficiencies such as over-distribution, leakages, and poor demand-supply alignment due to the absence of real-time responsiveness and predictive control. This study presents a machine learning (ML) framework for dynamic water supply regulation, integrating real-time sensor data with meteorological forecasts. The system was implemented and validated in Indore, Madhya Pradesh during March 2025, using data from smart flow meters, tank level sensors, and pressure gauges, along with temperature, humidity, and rainfall predictions. Two models—Random Forest (RF) and Long Short-Term Memory (LSTM) networks—were developed to forecast water demand over short-term and medium-term horizons, respectively. Predictions from these models were linked to an automated control system that dynamically managed valve operations, pump schedules, and leakage detection. The proposed ML framework achieved an 18% reduction in water loss, 14% energy savings in pumping operations, and a 20% improvement in demand-supply matching compared to traditional rule-based systems. Visualization dashboards and alert systems enabled proactive decision-making, while model performance metrics (R² and RMSE) confirmed the robustness of the predictive engine. This study demonstrates the viability of using ML-integrated control for municipal water supply, especially in rapidly urbanizing Indian cities, and lays the foundation for scalable smart water infrastructure.