Adaptive feedback linearization for an uncertain nonlinear system using support vector regression

This paper explores an adaptive feedback linearization for an uncertain nonlinear system using support vector regression (SVR). SVR, which assures global solution by quadratic programming (QP) problem, is used to learn the nominal dynamics of the input-output feedback-linearized system. Then, an adaptation algorithm of the offline-trained SVR is proposed for eliminating the offline-training error and uncertainties in the control process. In addition, the derivation of the adaptive rule considers the controller singularity problem by utilizing the affine property of the nonlinear system and the concept of the virtual control. Uniformly ultimately bound property of the overall system is analyzed by the Lyapunov stability theory. Simulations using a longitudinal dynamics of the F-16 model validate the performance of the proposed approach.