Nonlinear Predictive Control Based On Support Vector Machine Modeling

The results of traditional control methods such as PID control and LQ optimal control are not enough satisfactory when they are used for pure delay objects and non-minimum phase which often appear in the chemical industry,and they can do nothing when they are used for multivariable systems. While the predictive control algorithm just solves the problem, it has a very distinct advantage when used for the issues with operating variables of a high dimension; the issues that need to meet the physical constraints and delay systems. But by the development of the large-scale industrial processes, continuous increasing complexity of modern industrial processes, the linear predictive control methods just can’t well control these systems, and the nonlinear model predictive control are facing the large on-line computation problems. In this paper, we use SVM for modeling, which is linearised on-line taking into account the current state of the process; as a result,the nonlinear predictive control algorithm needs to solve a quadratic programming problem. In this way, one can reduce the calculation of the non-linear predictive control, and ensure the accuracy and stability of the control process.This thesis addressed the following several key problems:1. The content of prediction content and support vector machinesIn this paper, it has introduced the basic principle, the development process and the main characteristics of the predictive control. At the same time, described the basic theory of SVM, including comparing the different purposes for classification and regression.2. Nonlinear predictive control based on SVM Wiener modelsThe paper describes the SVM-Wiener model in detail, including the linear dynamic part and the nonlinear static part of model, designs a non-linear predictive controller and introduces the algorithm about local linearization of the nonlinear model predictive control.3. Nonlinear predictive control based on SVM Hammerstein-Wiener models.The paper briefly introduces the structural of the Hammerstein-Wiener model, a detailed derivation of the SVM modeling process based on Hammerstein-Wiener model is shown, and designs a nonlinear model predictive controller according to the structure and the characteristics. 4. Simulation results of the polymerization reactorIn the thesis, it compares the simulation results of MPC algorithms based on linear and SVM Wiener models. And the control accuracy and the computational efficiency of the algorithm are shown.In this paper, the number of all charts are29, including simulation diagrams and flowcharts, the number of all tables are8, including data comparison and defining the parameters, and there are also52references.