A fuzzy compensation mechanism in FFRLS-based adaptive MPC strategy

The control performance of Model Predictive Control(MPC) is strongly dependant on the quality of model, but system in reality more or less has time-varying properties, nonlinearities and un-modeled uncertainties. Therefore, an online adaptive model for MPC has been preferred in past years. This paper addressed a performance improving problem of Forgetting Factor Recursive Least Square(FFRLS) based adaptive MPC strategy. By identifying the distance between the current output and the expecting trajectory, the system´s state is classified, based on which two factors in control strategy(i.e. FF and weight of cost function) are fuzzily adjusted online. Moreover, an adaption stopping mechanism is also adopted to prevent the phenomena of estimator windup. Then the feasibility and superiority of the compensated controller is finally verified by simulation.