Switched Offline Multiple Model Predictive Control with Polyhedral Invariant Sets

This paper presents a switched offline multiple model predictive control procedure for nonlinear processes to ease the online computational burden and reduce the number of submodels. We employ the gap metric to characterize the dynamic difference between linear models and establish a linear model bank to approximate the nonlinear system. Based on the robust MPC algorithm, we develop an offline model predictive controller for each submodel. The polyhedral invariant set is utilized to expand the work scope of each local controller. In the offline part, a series of discrete states are selected, the corresponding feedback gains are precomputed, and associated polyhedral invariant sets are constructed. In the online implementation, the control input is simply calculated by calling the feedback gain according to the current state. A switching rule is then designed to integrate the submodels, and guarantee the stability of the whole system. Finally, the corresponding simulation example is presented to validate the efficiency of the presented algorithm.