Data assimilation for supporting optimum control in large-scale river networks

Abstract

We present a Nonlinear Model Predictive Control (NMPC) algorithm for real-time control of large-scale river networks in delta areas. The algorithm consists of an iterative, finite-horizon optimization of the system over a short-term control horizon. The underlying set of nonlinear internal process models represents relevant physical phenomena such as flow routing in the river network, and the dynamics of hydraulic structures.Data assimilation (DA) techniques turn out to be a key factor for the practical implementation of such schemes and may serve various purposes. First of all, DA contributes to the offline system identification of reduced internal models by parameter optimization. Secondly, we apply DA in an operational mode for model updating by adapting parameters, states, or outputs of the internal model for improving its lead time accuracy.The framework of DA and NMPC is applied on the control of a complex river network in the Dutch delta of Rhine River. We discuss the performance of a derivative-free optimization algorithm for calibrating the roughness coefficients of the underlying kinematic wave model and online parameter updating. Furthermore, we present the application of an Ensemble Kalman Filter (EKF) for updating model states as well as an output correction based on an AR(1) model. The contribution of these techniques in relation to the MPC performance is discussed in detail. Sx perf S perf J J P J J 978-1-4244-9573-3/11/$26.00 ©2011 IEEE

Towards integrating water prediction and control technology

“…This lead time can be effectively used to implement measures either to reduce the consequence of flooding through for example evacuation, or to reduce flooding itself through controlling dedicated hydraulic structures [9,29,31] or adhoc interventions, such as placing sandbags.…”

“…However, we would like to address general techniques which typically come along with models in operational applications. The most important ones are aiming at data assimilation [2,16,17,29] and uncertainty analysis. These techniques try to update the inputs, parameters, states, or outputs of a model based on historical observations for improving the lead time accuracy of the model or for providing information of the probable model error.…”

“…Non-structural measures in water systems, such as adjustable structures like gates, pumps, and sluices, are operated in order to keep water levels [9,11,29,31], flows [5,14,24,31], or water quality variables [1,17,18] in a desired state. There are many ways in which these adjustable structures can be controlled.…”

“…Several successful implementations of control systems on water systems are described in [21]. In increasing order of information usage, three possible types of control for water systems are feedback control [5,11,18], feed-forward control [17,31], and model predictive control [1,8,9,14,24,29].…”

“…In [3,15,26,30,29], model predictive control utilizing more accurate, hydraulic models ranging from kinematic wave models to full Saint-Venant based models are implemented. These non-linear models are still significantly reduced by using a large spatial discretization.…”