A Decision-Support System based on Real Time Control and Data Assimilation: A test case in Twentekanalen

The integration of forecasting and decision-making in real-time Decision-Support Systems (DSS) provides a powerful tool to operators of water resources systems for evaluating the future control of hydraulic structures. Decisions may be supported by presenting information about predicted disturbances, e.g. inflows into the water system, enabling the operator to try out future trajectories of structure control, or suggesting an optimum control based on predictive controllers. Ongoing work is undertaken under the programme Flood Control 2015 (FC2015) with respect to the management of flood events. This MSc thesis research was supervised jointly by the Operational Water Management research group of Delft University of Technology and the research institute Deltares. The aim of the MSc project is the transfer and extension of real-time DSS knowledge and techniques to a typical Dutch canal system such as Twentekanalen using simulation tools in development at Deltares. The main research objective is to assess the potential of DSS in this context and to investigate and verify a robust concept for applying Model Predictive Control on canal systems, taking into account missing or wrong data by applying Data Assimilation techniques. The main system characteristics and relevant processes of the Twentekanalen system are the following: \u95 3 Canals connected by locks in which the water level needs to be controlled. \u95 The water level is chiefly governed by the operation of locks, which need to turn in order for ships to pass, discharging a large quantity of water each time in comparison to other water flows in the system. Measurements of water level and flows at the locks are relatively complete. \u95 The water level is regulated by pumps and discharge structures at the locks \u95 Other water flows that occur in the system are lateral inflow and outflow. The measurements of these flows are relatively incomplete. At the start of the research a set of tools was available at Deltares. FEWS, a data management system, and RTC Tools, a reservoir routing model in development which was later extended with Data Assimilation capabilities. Near the end of the research a detailed model of the system in Sobek, a 1D and 2D water flow model, became available. A model framework has been designed to assess the potential of applying MPC and DA in a DSS for such a system. The incremental design and verification of this model framework has been the core of this research. The novel research is the addition of Data Assimilation techniques to Model Predictive Control. In order to show the added value of DA and verify its implementation a verification approach is needed to address the other components in the framework as well. The first method taken to achieve this was to set-up the MPC for Twentekanalen and integrate it into Delft-FEWS in hindcast mode assuming a perfect forecast. When the data set was made available it became clear that it contained large water balance errors. Adding DA showed improvements in the forecast, but while using realistic values for the DA, the forecasts were still far from accurate. By creating a workaround in the DA module it was shown that especially the Eefde-Delden reach had a large balance error that did not have a high correlation with the known lateral flows. Considering the low quality of the data set it was decided to expand the scope of the research and replace the data set by an accurate hydraulic model that became available near the end of this research. This model still uses measurements from the Twentekanalen system as input, but with internal controllers to regulate the pump and spill structures the water balance is maintained. With an extra expansion to inject known errors in the system, a thorough investigation of the effects of Data Assimilation and Model Predictive Control can be executed. First results from this expanded approach show promising results, but because of practical implementation issues of conflicting software modules, the full results will not be available within this research. Conclusions: From a theoretical point of view DA has a lot of potential. State updating solves an important issue of real time control; keeping the model state as close as possible to the real system state. Model training by parameter updating can be a good way to increase model forecasting performance. Online Parameter Updating can be very effective in systems were a high correlation occurs between measurements and unmeasured processes. These elements will make the model more robust, it can adapt to changing conditions. This also provides the model developer with interesting feedback on the workings of the modeled water system. From a practical point of view DA has shown improvements in the performance of the DSS as designed within this thesis project. But because of the large errors in the measurements it is difficult to translate these improvements to the effects in other systems. Implementation of the designed model framework gives a more satisfactory answer to that question. Recommendations have been made for improvements of the RTC Tools module, the development of prediction modules for the Twentekanalen system and further research using the developed framework with the models, scripts and programs written for this research. Most importantly getting predictions and real-time measurements on lock turning in the Twentekanalen system, and increasing the flexibility of model design in RTC Tools.Water ResourcesWatermanagementCivil Engineering and Geoscience