Water distribution systems (WDSs) must simultaneously satisfy consumer demand, maintain adequate pressure, and keep storage levels within operational bounds; objectives that require active adjustment of pump operations in response to changing conditions. This study introduces a framework that computes pump commands directly from measured operational data, eliminating the need for hydraulic model construction or calibration. The approach is validated across both levels of WDS control: at the device level, using the quadruple-tank process (QTP) as a laboratory-scale hydraulic benchmark in simulation, physical hardware, and real-time execution; and at the operational level, on a small-scale WDS under time-varying consumer demands. Across all case studies, the controller maintains tank levels and junction pressures within prescribed service bounds with mean absolute percentage errors below 4%, while using only 200 data samples. This is over 99% less data required by model predictive control (MPC). On the studied small-scale WDS, the controller achieves uninterrupted demand satisfaction, safe storage levels, and adequate junction pressures while reducing computational cost by 90% compared to a typical MPC. These results demonstrate the potential of data-driven, model-free approaches as a practical and data-efficient alternative for real-time control (RTC) of modern WDSs.
Villacres et al. (Mon,) studied this question.