Adaptive Control of Nonlinear System Based on SVM Online Algorithm

摘要

The training of Support Vector Machine (SVM) is an optimization problem of quadratic programming which can not be applied to the online training in real time applications or time-variant data source. The online algorithms proposed by other researchers are with high computational complexity and slow training speed. This manuscript combines the projection gradient and adaptive natural gradient. It proposes the constraint projection adaptive natural gradient online algorithm for SVM regression. An adaptive SVM controller is designed in the state feedback control for a class nonlinear system. In order to demonstrate the availability of this adaptive SVM controller, we give a simulation of the simple nonlinear system. The results of simulation demonstrate this SVM online algorithm controller is very effective and the SVM controller can achieve a satisfactory performance.