High-Performance Integrated Control of water quality and quantity in urban water reservoirs by dynamic emulation and model predictive control

Abstract:

Retention basins and urban reservoirs are increasingly used to support drinking water supply in large metropolitan contexts, since they make use of a resource, i.e., stormwater, that would be otherwise wasted, thus limiting the amount of water extracted from natural systems or produced with energy-intensive techniques. Yet, the operation of these infrastructures faces a twofold challenge. First, the presence of large impervious areas in urban catchments results in high discharge peaks and runoff volumes and a fast runoff response to rainfall, with consequent very short times of concentration. Second, stormwater transports large amount of pollutants to the receiving water bodies. This paper contributes a novel High-Performance Integrated Control framework to support the real-time operation of urban water supply storages affected by water quality problems. We use a 3D hydrodynamic, high-fidelity, simulation model to predict the main water quality dynamics and inform a real-time controller based on Model Predictive Control. We integrate the simulation model into the control scheme by a model reduction process, where the high-fidelity simulator is first used to identify and then replaced by a low-order dynamic emulator, which runs orders of magnitude faster. The framework is used to design the hourly operation of Marina Reservoir, a 3.2 Mm³ stormwater-fed reservoir located in the centre of Singapore operated for drinking water supply and flood control. Because of its recent formation from a former estuary, the reservoir suffers from high salinity levels, whose dynamics is modelled with Delft3D-FLOW. Results show that the real-time operation designed by our framework drops the minimum salinity levels of nearly 30% while reducing the average annual deficit of drinking water supply by about two times the active storage of the reservoir. Such a win-win solution is obtained by means of a model reduction process that reduced the dimensionality of Delft3D-FLOW by three orders of magnitude.