A machine learning-based surrogate model for coupled hydraulic and water quality simulation in water distribution networks

Ensuring consistent and high-level water quality is paramount for water utilities to meet health requirements and attain customer satisfaction. To this end, water utilities need to constantly surveil all relevant parameters such as chlorine concentrations and pH levels to optimally control water age, chlorine dosage rates, and other control variables in their water distribution networks (WDNs). While sampling experimental field data with high frequency and spatial density is cost- and time-intensive, water utilities need fast and accurate prediction capabilities for the relevant water quality parameters. Mathematical models coupling WDN hydraulic and water quality simulation have been well established in the literature and proven highly accurate. However, they are computationally very expensive such that optimization of control variables may only be possible to a very limited extent, especially for large WDNs. In this work, we propose the use of a lightweight, machine learning-based surrogate model for the prediction of water quality parameters that may serve to reduce simulation times and render optimization of control parameters more efficient. The baseline model is system-specific and learns to predict system state changes, i.e., the change in concentration or water age, based on common inputs to a hydraulic model such as water demands and actuator settings, i.e., pump speed and valve setting. In a next step, the model is targeted to be generalized for different WDNs by extending the input features to include network-specific topological measures, e.g., node centrality, node connectivity, and similar relevant graph theory measures.