Graph Neural Networks for Chlorine Prediction in Water Distribution Systems

<p dir="ltr">Chlorine concentration prediction in water distribution networks (WDNs) is critical for ensuring safe water quality and effective disinfection. This study proposes a surrogate modeling approach that combines physics-based simulations using EPANET/WNTR with a Graph Neural Network (GNN) for chlorine forecasting. The model captures both spatial dependencies and temporal dynamics of Chlorine by combining GNN layers with Gated Recurrent Units (GRUs). Synthetic data generated from the C-town benchmark network is used for training under diverse scenarios. The results demonstrate the developed model’s ability to predict chlorine concentrations accurately, supporting fast scenario exploration and practical decision-making.</p><p dir="ltr">This paper was presented at the 21st Computing and Control in the Water Industry Conference (CCWI 2025) at the University of Sheffield (1st - 3rd September 2025).</p>