Adaptive control vector parameterization for nonlinear model-predictive control

Chemical processes are often operated in a dynamic mode. This is always true by definition for the wide class of batch processes and it also holds for the transient phases of continuous processes, caused for example by load or grade changes. Nonlinear model-predictive control (NMPC) is a powerful approach to deal with the complexity of the related nonlinear control problems aiming for economic optimization. One major element of any NMPC implementation is a dynamic optimization algorithm, which solves the underlying optimization problem efficiently to obtain the optimal control trajectory in real time. Applying nonlinear dynamic optimization to industrial process models results in high calculation times restricting the use of NMPC for most chemical processes. In a previous publication, a method has been presented, which provides the optimizer with a control grid specifically tailored to the problem and therefore facilitates the optimization. In this work, this method is adapted to the on-line application in NMPC. In contrast to offline optimizations, uncertainties and a fixed discrete time sampling have to be taken into account. The approach is applied to a small example and considerable speed-up is observed. The gain is expected to be even greater for large-scale industrial problems. Copyright (c) 2007 John Wiley & Sons, Ltd.