Subspace Predictive Control Tuning with Multiobjetive Optimization

In this paper, a tuning method for data-driven, subspace predictive controllers is proposed. The tuning approach is based on the solution of a multiobjective optimization, in which the optimization problem is defined as the minimization of the quadratic error between closed-loop response and some desired reference trajectories. In its turn, these trajectories are described in function of user-defined time-domain objectives. The tuning parameters are obtained as the compromise solutions of the multiobjective optimization problem. Design choices and performance are discussed and the method is validated in computational simulations of some common types of process. Additionally, the tuning methodology is implemented in a multivariable pilot-scale thermoelectrical plant.