Implementation of Graph Neural Networks for Real-time Water Quality Forecasting in Water Distribution Networks

Abstract Submission: Real-time modeling and control of water distribution networks contribute significantly to the efficient management of water resources in addition to ensuring the required water quality is achieved. Due to the consequences of the use of disinfectants, adequate doses are necessary to ensure water quality. Therefore, a predictive approach potentially reduces vulnerability. In particular, water quality prediction has been based on using information from specific, but limited, hydraulic and water quality sensors to predict the quality at specific nodes. This study proposes to predict the chlorine concentrations in all network nodes through the application of a deep learning method. For this purpose, a Graph Neural Network (GNN) is implemented, considering information from both nodes and links (network topological characteristics, hydraulic, and water quality data). To evaluate the model, a small and a large network are used as case studies. The methodology of this work is based on (i) hydraulic and water quality simulation of the case studies, (ii) analysis and processing of the variables to be included in the model, and (iii) implementation of the GNN model. The purpose of this research is to demonstrate the predictive capacity of a GNN to know the information of the whole network. In this way, the operation of the systems can be more secure and efficient.

Learning Objectives/Expected Outcome (Optional) :