The Study Of Generalized Predictive Control Algorithm Base On Support Vector Machine Inverse System

In recent years, the inverse system method has established a nearlycomplete theory in the general form of nonlinear systems. This methodrequires a precise mathematical model of nonlinear systems, but thenonlinear characteristic of the actual project is often difficult to describeaccurately. Although established a complex mathematical model of thenonlinear system, the use of these complex models is also difficult to findthe analytical solution of the inverse model.Based on the statistical learning theory of the Support Vector Machine(SVM) has the advantages of good generalization capability and simplestructure. The characteristic of fitting function accurately has simplified thedifficult problem of modeling nonlinear system. The principle of SVM istransformed the difficult solution of nonlinear regression problem into a highdimensional feature space to solving a problem of convex optimization.Then the solution obtained a globally optimal.However, when the training samples have increased, the SVM would befaced with the curse of dimensionality and could not be training effectively.The Least Squares Support Vector Machine (LS-SVM) has overcomes theproblem of dimensionality which is a classic quadratic programming methodof solving support vector machine. It has good robustness, simple operationand fast convergence rate, therefore, it has more advantages in the area ofnonlinear control.Therefore, the paper combines the excellent recognition ability ofLS-SVM with the linear system method of traditional inverse system. First,use the LS-SVM identification inverse system, then connect the inversemodel in series with the original system, make up a pseudo-linearcomposite system to control. Second, this article describes the α orderinverse system method which based on support vector machine of SISOdiscrete-time nonlinear systems, and extended the method from SISO system to MIMO systems. Finally, use the simulation examples to investigate therecognition ability of LS-SVM and robustness of the open-loop system,which laid the foundation of in-depth studying and promoting theapplication of the method.In order to reduce the inverse system modeling errors, improve therobustness of the traditional inverse system and anti-jamming capability, inthe method of combine the inverse system method with generalizedpredictive control has proposed a new method of nonlinear generalizedpredictive control, which based on least squares support vector machine αorder inverse system method, do the closed-loop optimal control ofcomposite systems. The simulation results demonstrated that the proposedmethod does not rely on the precise mathematical model of the system, it hassuppression the system disturbances well, solved the nonlinear couplingproblems and improved the control effect of the nonlinear system.