Leveraging high-quality sensor data to enable simplified model-based digital twins for chlorine estimation

A model-based digital twin can support operational water-quality management in water distribution systems, but its practical deployment is often limited by solver fragility during repeated extended-period simulations. This study proposes an abstraction-based workflow to enable robust chlorine residual estimation using hydraulic models that are difficult to simulate reliably and may fail to converge. The method localises the network to target sensors, removes instability-driving elements such as pumps, valves, and tank regions, and reconstructs a simplified equivalent model while preserving dominant transport pathways. Flow directionality is enforced using check valves on high-flow links where flow direction is consistent. The abstracted model is then calibrated to real-time sensor measurements, including metered flows to define regional demand patterns and chlorine sensors to define time-varying injection setpoints. Using the observed chlorine and flow data, effective first-order decay rates for specific flow paths can be defined. Results on a full-scale case study demonstrate that this method results in high prediction accuracy for chlorine residuals (mean absolute error from 0.04–0.10 mg free chlorine/L). Significant variations in effective decay constants between distinct flow paths and over time highlight the need for path-based parameterisation rather than system-wide variables to estimate water quality.