Semi-parametric Adaptive Control Based On Extreme Learning Machine

The traditional methods have some limitations such as designing,interpretation and implementation of control systems for simultaneously solving parametric and non-parametric uncertainties,especially in modern digital discrete-time control systems.Semi-parametric adaptive control gives a novel theoretical framework to solve the plants with these two kinds of uncertainties.However,the existing semi-parametric adaptive estimation and control still have some problems,such as accuracy and real-time properties,which need to be further improved.This thesis systematically discusses the designing methods of semi-parametric modeling,estimation and control algorithms,and brings the concepts of computational geometry,extreme learning machine and data-driven into the design of controllers.In addition,this series of methods proposed in this thesis have been applied in robotic manipulator and motor control as useful attempts for practical problems.First of all,this thesis studies the characteristics of uncertainties existing in the general system and comparatively analyzes the classical control methods such as self-tuning regulator,model reference adaptive,robust control,model-free adaptive and so on.A novel kind of method called semi-parametric modeling and analytical method is introduced,which highlights the respective characteristics of parametric uncertainty and non-parametric uncertainty.The mathematical descriptions of priori information are established for typical systems so that information concentration estimation algorithm is used to solve the estimation problem of parametric uncertainty part.Second,the main steps and basic problems of the information concentration estimation algorithm in the case of two-dimensional parameters are described.The geometric relationships between lines and polygons in the information concentration process are analyzed in detail.At the same time,the algorithm flow chart of information concentration transformation is designed.The simulation examples of the proposed information concentration estimation algorithm are depicted.The computational complexity optimization of information concentration estimation has been discussed.What’s more,the main advantages and disadvantages of information concentration estimation have been summarized.Third,the mathematical descriptions of semi-parametric adaptive trajectory tracking are given,and the problem of adaptive estimation and control of semi-parametric model is analyzed.The algorithm characteristics of the extreme learning machine are analyzed.The non-parametric part estimation method based on the appropriate variant algorithm of extreme learning machine has been proposed.According the adaptive estimation of information concentration and extreme learning machine,one kind of adaptive controller is designed for the trajectory tracking problem of semi-parametric systems.The simulation examples are used to test and verify the performance of the proposed control algorithms.Finally,the control difficulties of robot trajectory tracking and servo motor motion control with multiple uncertainties are analyzed.Semi-parametric adaptive estimation and control based on extreme learning machine are applied to solve the motor control in robot motion problem.The performance of proposed adaptive motion control has been tested and verified through simulation comparison experiments.In the end,the main contents of this thesis are summarized and the future research directions of this topic are discussed.