An anticipatory fuzzy logic controller utilizing neural net prediction

Abstract

The goal of this project was to evaluate control fuzzy logic for applicability to con trol of flexible structures. This was done by applying these methods to control of the Control Structures Interaction Suitcase Demonstrator developed at Marshall Space Flight Center. Both traditional and new anticipatory fuzzy logic schemes were applied to the system, and results were compared to that of the system with a standard Linear Quadratic Regulator as a controller. In order to perform the state prediction necessary to the anticipatory fuzzy logic controller, a neural network was trained to emulate the behavior of the system, based on input-output data for the system. Behavior of the controllers was compared under ideal conditions, under noisy conditions, and with randomly chosen state parameters perturbed by + or-50%. Fuzzy systems demonstrated robustness to added noise and to changes in plant parameters; the anticipatory fuzzy system exhibited superior performance when compared to both traditional fuzzy and LQR controller systems. The anticipatory fuzzy neural controller exhibits similar properties, but does not require that any mathematical model for the system exist. Thus it can be applied to many real world systems for which other control methods can not be used.