Hardware-in-the-Loop Validation of Data-Driven Predictive Control for Optimal Water System Operation

Model predictive control (MPC) is recognized for optimizing operations in water distribution systems (WDSs) by handling constraints and predicting system behavior. However, creating accurate system models is challenging for large WDSs with complex layouts and numerous hydraulic components. This paper introduces a model-free predictive control framework using neural state-space (NSS) modeling, which employs neural ordinary differential equations to identify system dynamics without physics-based models. The proposed NSS-MPC framework is experimentally validated through hardware-in-the-loop (HIL) testing on a quadruple tank system (QTS) with pumps, pipes, valves, and sensors. HIL testing integrates physical hardware with control software, offering a realistic, cost-effective evaluation platform. With SCADA integration, the framework supports real-time monitoring and system visualization. Case studies demonstrate NSS-MPC’s ability to outperform traditional controllers, such as proportional integrators (PI), by achieving smoother control and better tank pressure head management. The results highlight the scalability and practicality of NSS-MPC for improving WDS operations.