The Research And Application Of Nonlinear GPC Based On SVM

With the successful application in the industrial process control, Generalized Predictive Control (GPC) algorithm has attracted increasing attention in fields of control theory and engineering for its property of robustness and ability to resist disturbance. The GPC algorithm is based on linear model. However, practical industrial processes usually are always complex with multiple inputs multiple outputs, strongly nonlinear, strongly coupling features. The GPC algorithm that based on Controlled Auto-Regressive Integrated Moving Average (CARIMA) model can hardly been applied directly. To these cases, support Vector Machine (SVM), which is based on statistical learning theory, is one kind of new machine learning algorithm. Because of its good nonlinear function approximation capability and generalization ability, it has been widely applied in system identification and other control areas. Therefore, this paper tries to combine the SVM system identification method with GPC algorithm, make efforts to present a user-friendly method of generalized predictive control of nonlinear systems, Specific research contents are as follows:1. For the strongly nonlinear system, firstly, its model is identified by SVM with the kernel as Radial Basis Function (RBF), and then, linearization the SVM model using Taylor expansion Method, finally, adjust the linear model to meet CARIMA model form, the GPC controller will be designed based on this CARIMA model. To deal with the instability problem that is caused by model mismatch when linearization of the SVM model, an additional penalty factorμis introduced to improve the system's stability, at the expense of the cost of the optimal controller.2. In order to solve the problem that the linear SVM model method is very difficult to apply to strong coupled MIMO coupling system, inverse control method which is widely used in nonlinear system control area is introduced in the paper. Using SVM to identify the a -order inverse model of the original controlled system, the inverse model then is cascaded before the original system, for single-input single-output (SISO) system that will create anα-order pure delay pseudo-linear sub-system, for multiple-input multiple-output (MIMO) coupling system that will create several independentα-order pure delay pseudo-linear sub-systems. Because theα-order pure delay pseudo-linear sub-systems meet the CARIMA model, then the GPC controller for the MIMO coupling system can be easily designed, simulation results verify the effectiveness of the algorithm.3. Implementation the GPC algorithm based on support SVM inverse control to an actual model of a paper machine, to achieve the decoupling control of the two important parameters of paper quality: basis weight and moisture. Simulation results show the effectiveness of the algorithm proposed in paper.