A Smart Water Distribution Network Management Framework based on Internet of Things and Reinforcement Learning Technologies

Abstract

The convergence of the Internet of Things (IoT) and Reinforcement Learning (RL) has transformed the operation of urban water distribution networks (WDNs) from manual, reactive supervision to intelligent, self-optimizing control. This paper presents a hybrid IoT–RL framework that integrates real-time sensing, data synchronization, adaptive control, and predictive decision-making for efficient network management. The proposed architecture employs distributed IoT sensor nodes to continuously acquire pressure, flow, and energy data, which are transmitted via MQTT to a time-series database for feature extraction and anomaly detection. A Proximal Policy Optimization (PPO)–based RL agent learns dynamic control strategies that minimize leakage losses and energy consumption while ensuring hydraulic stability. The framework is evaluated in the EPANET 2.2 environment using the C-Town benchmark under 24-hour diurnal demand conditions. Three control configurations are strictly studied: (i) the SCADA (Model A), which is a static threshold control, (ii) the hybrid PID-Model Predictive Control (PID-MPC) controller (Model C) as a state-of-the-art non-RL control, and finally (iii) the proposed adaptive IoT-RL control (Model B). Quantitative measures show that the performance of Model B is both 93, 22 improved in decision latency and control accuracy compared to Model A and has a 12, 15-percent better latency and accuracy improvement over the sophisticated Model C. Quanti-tative metrics including control accuracy, latency, data throughput, and network efficiency performance index (NEPI) demonstrate that Model B improves decision latency by 93 %, control accuracy by 22 %, and pro-cessing throughput by 95 % relative to Model A, while maintaining superior reliability over Model C. All fig-ures were programmatically generated in MATLAB (vector export) using EPANET-based topological and con-trol visualization scripts. The results confirm that the IoT–RL integration provides a scalable, data-driven pathway for sustainable water infrastructure management and proactive leakage mitigation.