Study On Support Vector Machine Inverse System Method And Its Application

In recent years, a relatively integrated design theory of inverse system method has been constructed for the general nonlinear system. However, as is known to all, this method is based on an accurate mathematical model of the controlled system. In many practical cases, however, the accurate mathematical description of the system is almost unknown. Even if the original system's model is completely known in advance, the nonlinearities and complex coupling relationships contained in the model still make it quite hard to deduce the inversion. To deal with this problem, firstly, the Support Vector Machine (SVM) inverse control of SISO nonlinear discrete systems is introduced, and the existence condition of SVM a th-order inversion is proved in theory. And also the SVM inverse control of SISO nonlinear discrete systems is extended from discrete systems to continuous systems and from SISO systems to MIMO systems. All these lie a foundation for further studying and extending applications of SVM a th-order inverse system method. Secondly, to overcome the problem of low speed of SVM, the paper presents a Least Squares Support Vector Machine (LS-SVM) a th-order inverse system method for nonlinear system. Then the LS-SVM a th-order inversion based predictive control is proposed. To improve the robustness and ability to resist disturbance of the traditional inverse system method, the paper combines the inverse system method and Internal Model Control (IMC) to present a novel method: the nonlinear internal model control based on LS-SVM a th-order inverse system method. Finally, the theoretical method is used in the Air Dense Medium Fluidized Bed (ADMFB) control, and gets good simulation results. The main contents of the paper and the achievements are as follows:Firstly, to conquer the two bottlenecks of the traditional inverse system methods, combining the good ability of modeling nonlinear processes under small data set available of SVM, the SVM inverse control of SISO nonlinear discrete systems is studied, and the existence condition of SVM a th-order inversion is proved in theory. And also the SVM inverse control of SISO nonlinear discrete systems is extended from discrete systems to continuous systems and from SISO systems to MIMO systems. Theoretical analysis and simulation results show that the method has satisfactory performances and doesn't depend on an accurate mathematical model. This approach is a novel method available for nonlinear systems.In the second part, LS-SVM a th order inverse system method for nonlinear systems is presented to deal with the problem of low speed of SVM. The existence condition of SISO and MIMO SVM a th-order inversion are given respectively. The better performance of this method over SVM a th order inverse system method isvalidated by some simulations. So this method is of more significance from both theoretical and practical point of view.In the third part, LS-SVM a th-order inversion based predictive control is proposed. The main idea of this method is that it takes the pseudo-linear system as its controlled subject, which is well controlled by introducing the predictive control strategy. This method doesn't need an accurate mathematical model of the nonlinear system and has the characteristics of good control performance, strong anti-disturbance ability, and is simpler to use. The design of the predictive controller is greatly reduced by the method proposed in this paper.Nonlinear internal model control based on LS-SVM a th-order inversion is presented in the fourth part. To improve the ability of robustness and anti-disturbance of the traditional inverse system method, a new internal model control method based on LS-SVM a th-inversion is proposed. The internal model control method is introduced to the pseudo-linear system, which is gotten by LS-SVM a th-order inverse system method. Both the theoretical analysis and the simulation results show that the combined method doesn't depend on an accurate mathematical model and has the characteristics of better robust stability, simply use and higher tracking accuracy. And this approach is one of the methods f available or nonlinear systems.Fifth, the application of nonlinear internal model control based on LS-SVM ath-order inversion in ADMFB is studied. Because the dry cleaning of coal with ADMFB is a multi-variable, nonlinear, strong coupling and complex process, traditional control methods aren't able to control it effectively. To control ADMFB effectively, the technologic process is analyzed and the mechanism model is deduced based on simply mechanism analysis. The invertibility of the system is proved and the nonlinear offline a th-order inverse model of the control plant is built by LS-SVM. Combining LS-SVM ath-order inverse system with the original system can form a pseudo-linear system, which is linearized and decoupled. Then the internal model control method is introduced to the pseudo-linear system. The effectiveness of the method is validated by simulations.