Tuning of Model Predictive Control Based on Hybrid Optimization

Abstract

Abstract A multi-objective optimization methodology is presented to find optimal tuning values of model predictive controllers (MPC). The method uses an optimization layer above the controller that provides the prediction and control horizons and the weighting values of the objective function. The methodology uses a hybrid method: the goal attainment method and the variable neighborhood search. The first one allows minimizing the error between the closed-loop response and an output reference trajectory varying the weighting matrices until reaching the minimum multi-objective cost, while the second method varies the integer variables of the problem (prediction and control horizons) until finding the minimum horizons to obtain a desired closed-loop performance. The proposed method is tested in a benchmark process using a Generalized Predictive Controller (GPC), showing satisfactory results.