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 water quality parameters, for example, chlorine-concentration, as well as to optimally control dosage rates in their drinking water distribution systems (DWDSs). Simulation models coupling DWDS hydraulics and water quality have been well established and highly accurate. However, they are computationally very expensive such that optimization of control parameters may only be possible to a very limited extent. In this work, we are proposing the use of a lightweight, machine learning-based surrogate model for the coupled simulation of hydraulic and 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 the steady-state hydraulic and water quality state simultaneously based on common inputs to a DWDS model, that is, water demands and dosage rates at reservoir levels. Results indicate good prediction capabilities of the surrogate model with R² values greater than 0.98 as well as error rates below 0.01% for hydraulic parameters and below 1% for water quality parameters. Some slight spatial trends in the prediction error’s variance are identified for hydraulics as well as for water quality parameters.