Incorporation of the Generalized Tsk Models in Model Predictive Control

Abstract

The generalized TSK (GTSK) modeling approach is proved to provide accurate model prediction and to alleviate the computational burden. The scope of this study is to incorporate the GTSK models in the nonlinear model predictive control (NMPC) to improve the overall performance and reliability of NMPC. A novel global optimization method, the Leapfrogging technique, is also used to further improve the NMPC's computational efficiency. Another innovation, the "sawtooth" pattern is used as input signal to generate the GTSK model. The experiments and tests are conducted on a nonlinear process simulation system, in which the NMPC control algorithm was embedded. The virtual process in this simulator is fourth-order-plus-dead-time (FOPDT) process with a nonlinear gain and the environmental effect (noise and disturbance). The controlled process is subject to both soft and hard constraints - soft on both the controlled and the auxiliary variable, and hard on both the limits and rate of change of the manipulated variable. The NMPC performance is evaluated via several simulation experiments, which involved constraint handling, interactions and process nonlinearity. The use of a GTSK model and Leapfrogging as an optimizer were demonstrated as effective for nonlinear model predictive control. The nonlinear model is firstly developed by using GTSK approach. The prediction accuracy of the GTSK model was illustrated and quantified by a comparison with SOPDT model. The GTSK model was much better. The performance of GTSK MPC controller is evaluated via seven sets of dynamic control simulation. The controller showed desirable performance for disturbance rejection, set point tracking, constraint handling, and comprehensive environmental effect handling.