Adaptive Sliding Mode Control based on SVR

Abstract

Sliding mode control (SMC) is a prevalent control technique, especially effective for nonlinear systems. Its performance is enhanced if the parameters chosen in the design of the SMC are determined in an optimal way. In this paper an SMC architecture is implemented where support vector regression (SVR) methodology is employed in optimizing one of the design parameters of SMC. A major strength of SVR with respect to gradient based optimization methods is that it finds the global minimum by formulating a convex cost function. The proposed control architecture is tested by simulations performed on an inverted pendulum system. Also, the robustness of the method is justified by additional simulations with measurement noise and disturbance added to the system.