Automatic Tuning of Model Predictive Controllers Based on Fuzzy Logic

This paper presents an online tuning strategy for model predictive control. Specifically, the tuning strategy adjusts automatically the prediction horizon, P, the diagonal elements of the input weight matrix, Λ, and the diagonal elements of the output weight matrix, Γ. The control horizon is left constant because its relative value with respect to P is more important. The MPC parameters are adjusted such that the resulted feedback response satisfies certain time-domain performance specification. The tuning algorithm is based on fuzzy logic. Predefined fuzzy rules that formulate the general tuning guidelines available in the literature and the performance violation measure in the form of fuzzy sets determine the new tuning parameter values. Therefore, the tuning algorithm is cast as a simple and straightforward mechanism with modest computational requirements. The effectiveness of the proposed tuning method is tested through simulated implementation on an evaporator example.