Applications Oriented Input Design for MPC: An analysis of a quadruple water tank process

Model predictive control (MPC) has become an increasingly popular control strategy thanks to its ability to handle multivariable systems and constraints. This control technique makes use of a model of the system, therefore performances are highly dependent on the accuracy of the model chosen. The process of obtaining the model is often costly, for this reason system identification for MPC is an important topic. Applications oriented optimal input design enables optimization of the system identification experiments, leading to a set of models with the necessary accuracy for the intended application. In this thesis a method of system identification for MPC applications is simulated on a multivariable nonlinear system consisting of four interconnected water tanks. An analysis of the impact of MPC settings, such as active or no active constraints and different weight settings, is carried out. The initial hypothesis is that different MPC settings influence the obtained set of models. Simulations show that the hypothesis is correct and the result give rise to some interesting interpretations.