No abstract

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 3‐D, high‐fidelity simulation model to predict the main water quality dynamics and inform a real‐time controller based on Model Predictive Control. The integration of the simulation model into the control scheme is performed by a model reduction process that identifies a low‐order, dynamic emulator running 4 orders of magnitude faster. The model reduction, which relies on a semiautomatic procedural approach integrating time series clustering and variable selection algorithms, generates a compact and physically meaningful emulator that can be coupled with the controller. The framework is used to design the hourly operation of Marina Reservoir, a 3.2 Mm3 storm‐water‐fed reservoir located in the center 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 behavior is modeled with Delft3D‐FLOW. Results show that our control framework reduces the minimum salinity levels by nearly 40% and cuts the average annual deficit of drinking water supply by about 2 times the active storage of the reservoir (about 4% of the total annual demand).

No abstract

No abstract

Marina Reservoir is one of the largest fresh water body in Singapore, recently constructed with the purpose of increasing the drinking water supply and Singapore's self-sufficiency. Besides this strategic role, the reservoir also serves for floods control and lifestyle attraction. The largest portion of the inflow volumes comes from five main uncontrolled catchments, which show fast-varying discharge variations due to the strong storms affecting this region. Moreover, highly paved areas cause rapid runoff processes, with a concentration time lower than one hour. Such hydrological context creates difficulties in satisfying the operational objectives and provides a challenging environment for the application of real-time control methods. With the purpose of evaluating the potential of real-time control for the optimal operation of Marina Reservoir, this work adopts Model Predictive Control (MPC), a form of deterministic control that employs the current state of the system, the future inflows trajectory furnished by a predictive model and a further model describing the internal dynamics of the controlled subsystem to determine an optimal control sequence over a finite prediction horizon. The rationale behind the choice of MPC is that this method is characterized by reduced computational requests and the capability of exploiting any exogenous information, as precipitation or evaporation measurements, that allow for an accurate prediction of the inflow events, which is crucial for an effective management of fast-varying hydrological systems. In the present application, Marina Reservoir catchment is described with a data-driven rainfall-runoff model (in the form of M5 model trees) that fully exploits the hydro-meteorological information available in real-time, thus enhancing the accuracy of the inflow prediction. The model is combined with a non-linear MPC scheme that optimizes the barrage operation according to the different operational objectives. Preliminary results show the effectiveness of the proposed control method, which outperforms the currently-used operating rules.