A Study On Variable Structure Control Based On Support Vector Machine

Support Vector Machine (SVM), which is based on Statistical Learning Theory (SLT), has drawn more and more attention in recent years because of its advantages to other learning machines. It has been applied in black box or gray-box systems' recognition and control now. Variable Structure Control (VSC) has been researched from 50's last century, but there are no researches on application of SVM in VSC. In order to utilize advantages of SVM and VSC, several kinds of VSC schemes based on SVM are studied in this dissertation.The main contributions of this dissertation are outlined as following:First, basic algorithm for single-input-single-output (SISO) system based on SVM is discussed. Then by adopting character of reducing dimension for VSC, basic VSC schemes for single-input-multi-output (SIMO) system and multi-input-multi-output (MIMO) system based on SVM are proposed.Second, on the basis of basic VSC algorithm for control system based on SVM, SVM is introduced in realizing feed-forward and VSC feedback technology. So effect of external disturbance is reduced and robustness of VSC system is guaranteed.Third, after simply introducing character of moving switching surface, a VSC scheme based on SVM and moving switching surface is brought forward. Moreover, in order to overcome shortages of moving switching surface, a VSC scheme based on SVM and combined switching surface is introduced to resolve contradiction between reaching speed and robustness when the moving switching surface is adopted.Forth, Some problems related to VSC control systems based onSVM are discussed. Trajectory-tracking condition of linear discrete-time system is proposed. Then existing condition of inverse model for VSC system is discussed. And effect of switching surface parameters on system performance is analyzed.At last, whole paper is summarized and open problems of VSC scheme based on SVM in theory and application are put forward.