Multinode Real-Time Control of Pressure in Water Distribution Networks via Model Predictive Control

Leakage represents a crucial issue in the management of water distribution networks (WDNs). Due to leakage dependence on service pressure, the application of pressure real-time control (RTC) can effectively alleviate the problem. Current RTC implementations rely on a closed-loop control of pressure at a single (local or remote) node of the WDN. In this case, the regulation performance may be good at the selected node, but rather poor across the whole WDN. While conservative choices are usually carried out to mitigate it, this issue becomes extremely relevant in the case of multiple nodes reaching critical values of pressure during the day. This work proposes a novel multinode (MN) RTC approach, which explicitly considers closed-loop control of pressure at multiple WDN nodes. The control scheme is based on a Kalman Filter for state and disturbance estimation, a steady-state auxiliary target calculator, and a model predictive controller for regulation and disturbance rejection. A detailed pressure-driven, unsteady flow model is used to simulate a real WDN under different demand scenarios and assess the performances of the proposed approach, which delivered satisfactory results. Moreover, the MN-RTC approach discussed in this work is suitable for in situ implementation, due to its low computational complexity. Finally, as demonstrated in the simulated environment, the tuning of the control algorithm can be performed by relying on input-output data collected directly from the plant, with no need for a hydraulic simulator.

Automated summary

Pressure real-time control is an effective way to address the issue of leakage in water distribution networks. However, current closed-loop control methods only focus on a single node, resulting in poor overall performance. This study proposes a novel multinode RTC approach that considers pressure control at multiple nodes using a Kalman filter, steady-state auxiliary target calculator, and model predictive controller. Simulation results demonstrate satisfactory performance and low computational complexity, making this approach suitable for practical implementation.