A Generalized Adaptive Neural Network Fuzzy Inference Structure for Nonlinear Control

摘要

The control of a nonlinear system is a challenging problem particularly when the system has some uncertainty or there are imperfections in the model dynamics. One approach that has gained some success employs a fuzzy structure in concert with a neural network (ANFIS); the fuzzy component compensates for the uncertainty while the neural network component models the underlying system dynamics. This paper presents a generalization of previous work in which several ANFIS blocks are employed to emulate PI control, PID control or other control structures. Assuming that a desired controller can be modeled by a differentiable function, this generalized ANFIS controller can approximate the control action. The advantage of using this structure is that it can provide the desired control action while compensating for uncertainty and while learning about the underlying model dynamics. An example of comparing the approach to other controllers is provided to illustrate the features of this method.