D.7 Development of toolbox for linear and non-linear MPC

The research concerning model predictive control focuses on development of toolboxes for both ICAS and Matlab. The model predictive controllers are based on a receding horizon principle in which the variables that can be manipulated are controlled at each sampling time by solving a quadratic or linear optimisation problem. The linear model predictive controller is in state space form and provides guaranteed nominal stability. A Kalman-filter is used for state estimation from the measurements. The linear models are identified by subspace identification. Stability properties are also investigated.
The model predictive control methodologies developed has been applied to a number of periodic operating processes, i.e. to control a simulated industrial flow swing reactor, and is being investigated for control of an activated sludge process.
Future work focuses around the non-linear model state estimation and predictive controller. Currently the methodology is based upon a first-principles model and either an extended Kalman-filter or a receding horizon estimator for estimation of the states given the outputs. Approaches for utilising the special structure of the optimisation problem are being investigated.