Evaluation of experiment designs for MIMO system identification by model predictive control

Ten different experiment designs for control-oriented MIMO identification are evaluated by the performance of model predictive control (MPC) using the identified models for control of the true system, which is a moderately ill-conditioned 3×3 system. The evaluated designs are some standard designs as well as advanced designs aimed at ensuring integral controllability requirements. Different types of input perturbations are also considered. Control performance in terms of output error and control activity measures is determined for various choices of sampling time and input constraints. The evaluation clearly shows that for this application, experiment designs where the directionality of the system is taken into account produce better models for control design than more standard methods.