Decoupling Generalized Predictive Control Using Multiple Models

Most of the plants in process industrial are multivariate coupling systems. The associations often exist among control variables and controlled variables. It's very important to decouple the system if you want to get the nice control effects. The Generalized Predictive Control (GPC) has nice control quality and robustness, but it doesn't consider the coupling of the control loops, so the tracking performance is not very good.Firstly, this paper presents three algorithms of decoupling GPC: The first algorithm is GPC algorithm with reference observer, which raises the errors'weight matrixes of the GPC's objective function. So the outputs will follow the setpoints and the decoupling will be weakened. The second algorithm is GPC algorithm with feed-forward decoupling. Regarding the other channels'impacts to the researching channel are disturbs, it can decouple the system. During it, the computing approximate equation is omitted and it is related with the outputs of controller. The third algorithm is GPC algorithm with objective function decoupling. It decomposes the system into multiple inputs and one output system, then uses GPC to design controller separately by means of solving matrix equation groups, later GPC decoupling controller is realized. It and reference observer can be used in GPC, Dynamic Matrix Control (DMC), and Model Algorithmic Control (MAC). The article also adds the reference observer into the GPC algorithm with feed-forward decoupling and the GPC algorithm with objective function decoupling, so the decoupling can be decreased more. The simulation results show effectiveness of the above algorithms.Secondly, for complex industrial processes which are nonlinear and strongly coupled, the decoupling GPC using multiple models algorithm is proposed. It gets many models at the equilibrium points through linearization, it then chooses the best model via the switching index. At last, the GPC algorithm with reference observer is used to control the system. The simulation results show good control effects.Finally, for the medium speed mill which is nonlinear, strongly coupled and wide working condition range, the system model is built and the decoupling GPC using multiple models algorithm is put forward. It combines multiple modes GPC, GPC algorithm with feed-forward decoupling, and GPC algorithm with objective function decoupling. The simulation examples show effectiveness of the proposed algorithm.