Flood control of the Demer by using Model Predictive Control Maarten Breckpot a,n , Oscar Mauricio Agudelo a , Pieter Meert b , Patrick Willems b , Bart De Moor a a KU Leuven Department of Electrical Engineering (ESAT-SCD)/iMinds Future Health Department, Kasteelpark Arenberg 10, B-3001 Leuven, Belgium b KU Leuven Department of Civil Engineering, Kasteelpark Arenberg 40, B-3001 Leuven, Belgium article info Article history: Received 28 February 2013 Accepted 16 August 2013 Keywords: Model Predictive Control Flood control Kalman filter Open channel flow abstract It is shown how Model Predictive Control can be used for flood control of river systems modelled with real data. A linear model for the Demer, a river in Belgium, is derived, which is used inside the optimisation problem solved by the controller. This optimisation problem is formulated such that the controller can be used for set-point and flood control. A Kalman filter is used as a state estimator. Closed loop simulations performed with a full hydrodynamic model of the Demer in combination with historical rainfall data show that the proposed control scheme outperforms the current control strategy. & 2013 Elsevier Ltd. All rights reserved. 1. Introduction River floods are worldwide a serious problem. One example of a river in Belgium with a long history of severe floods is the Demer. In order to reduce these floods, the local government has installed water reservoirs to store the excess of water during periods of heavy rainfall. Also hydraulic structures were built to control the discharges in the river and the water going to and coming from these reservoirs. Although these adaptations significantly reduced the risk of flooding along the Demer, extreme rainfalls in 1998 and 2002 still resulted in severe floods. Simulations performed by the government have shown that these floods could have been reduced or even avoided if a more advanced control strategy had been used. Previous works exist where Model Predictive Control (MPC) is successfully used for flood control and set-point control of the Demer based on a simple model. The goal of this research is to further improve these MPC schemes by working with an approximate model of a full hydrodynamic model of the Demer and test the controller on this white box model. MPC (Mayne, Rawlings, Rao, & Scokaert, 2000; Rossiter, 2003) originates from the chemical process industry but it has shown great value in many different applications going from food processing to automotive and aerospace applications (Qin & Badgwell, 2003). Because MPC formulates the control problem as an optimisation problem, it can be used to control rivers during different operating conditions. By minimising the deviations of water levels from their targets, MPC will focus on set-point control during periods of no or little rainfall. By incorporating the flood levels as constraints on the water levels in the optimisation problem, the same MPC will focus on flood control during periods of heavy rainfall. However only a high control performance can be achieved if an accurate model is used in the optimisation problem. The dynamics of a reach of a river can be modelled with the full hydrodynamic equations of de Saint-Venant, or the so-called Saint-Venant equations (Chow, Maidment, & Mays, 1988). Based on these equations together with the dynamics of hydraulic structures, junctions and reservoirs, a mathematical model can be derived for a river system. Because of the complexity of these models, MPC cannot work directly with these equations but use approximate models. Models derived by means of identification techniques or simple integrator delay models provide a good approximation and have been used in combination with MPC in many studies for set-point control (Puig et al., 2009; van Overloop, 2006; van Overloop, Clemmens, Strand, Wagemaker, & Bautista, 2010; Wahlin & Clemmens, 2006). However these models are not accurate enough for the purpose of flood control because they model the water levels at only one location. Therefore it is hard to guarantee that the water level profile along the entire reach is below the flood level. In previous works done by our research group a conceptual model was used to model the Demer (Barjas Blanco et al., 2010; Breckpot, Barjas Blanco, & De Moor, 2010; Barjas Blanco, 2010). However this model approximates the dynamics of every reach at only a very limited number of points. Because the flood levels are very irregular, it is never certain that Contents lists available at ScienceDirect journal homepage: www.elsevier.com/locate/conengprac Control Engineering Practice 0967-0661/$ - see front matter & 2013 Elsevier Ltd. All rights reserved. http://dx.doi.org/10.1016/j.conengprac.2013.08.008 n Corresponding author. Tel.: þ32 16328652; fax: þ32 16321970. E-mail addresses: maarten.breckpot@esat.kuleuven.be, maarten.breckpot@hotmail.com (M. Breckpot), mauricio.agudelo@esat.kuleuven.be (O.M. Agudelo), pieter.meert@bwk.kuleuven.be (P. Meert), patrick.willems@bwk.kuleuven.be (P. Willems), bart.demoor@esat.kuleuven.be (B. De Moor). Control Engineering Practice 21 (2013) 1776–1787