Performance Based Sliding Mode Controller Using Support Vector Machines

Abstract Abstract Sliding mode controller with time-varying sliding surfaces is a method to improve robustness and transient response of a system. In this study, a special time-varying mechanism is considered in which the sliding surface depends on angular information and sliding surface parameters are obtained with respect to the given initial conditions. Support vector machine regression algorithms are used in order to obtain the parameters of the time-varying sliding surface for any initial condition chosen from the predefined phase space interval. The support vector machine is trained for a number of initial conditions and computer simulations are presented to show the average improvement on randomly chosen test data with respect to the conventional sliding mode controller and to the time-varying sliding mode controller for which the parameters are tuned with a genetic algorithm for each initial condition. It is seen that, by using support vector machine based parameter tuning for different initial conditions, both the transient response of the system and reaching time is improved with respect to the conventional sliding mode controller and similar performance indices can be obtained with the controller in which parameters are tuned with a genetic algorithm having a computational time burden.