Dynamic Modeling And Intelligent Control Of The Constrained Flexible-Link Robot Manipulators

In order to deal with problems such as non linearity and nondetermination,the intelligent control that focus on fuzzy logic technology and neural network theory has been becoming more and more emphasized since 80's.Neural networks and fuzzy systems are essential nonlinear systems and capable of approximating to almost all functions and their derivatives of every order.Therefore,as a kind of identifier independent from system models,they may describe nonlinear control systems with generalized mathematics models .On the other hand,with powerful ability of synthesis infonnation,they can largely and simultaneously process different kind of inputs to unravel the problems of complementary and redundancy among inputs.Meanwhile,their parallel structure s include potentially the advantage in real time control systems.It is a lot of favorable charactinics that neural networks and fuzzy systems have been widely used in many fields.Many successes in application have been made in recent ten years,but many new problems also show up.Firstly.Through real time is a basic requirement for automatic control system,generally the convergence speed of learning algorithm of neural networks is too slow to satisfied the requirment.Essentially,the parameter learning procedure of neural networks is nonlinear optimation.At presnt,the learning algorithm for the modeling and control of nonlinear systems still base on gradient ,and this usually resulted in the local minmum.Secondly,it has been verified theoretically that sufficient hidden nodes can satisfy the approximating accracy,but how many hidden nodes needed to guarantee the desired accuracy remained unsatisfied. Moreover,super-scale neural networks are difficult to realize and delay the learning period.So,it is a practice problem how to guarantee the approximating accuracy within a certain scale of networks.Based on analysis mention above,several researches are developed here as follow:First, a generalized expression for neural networks is given after comparing their structure.No matter neural network or fuzzy neural network,their mapping relation can be described as the linear combination of a series of variable basis functions,which show the functions of hidden layer-the number of the hidden nodes is related to that of the variable basis functions,the form...