A NEURAL NETWORK SATELLITE ATTITUDE CONTROLLER WITH ERROR BASED REFERENCE TRAJECTORY Valdemir Carrara

Feedforward neural networks are investigated in this work to verify its ability to control the attitude of a satellite. Neural nets are a promising tool for attitude control due to its inherent nonlinear behavior, which makes them a natural candidate to control nonlinear systems. Nevertheless, as will be shown, to obtain a neural 3 axis attitude control is not as simple as a conventional SISO net control. Main difficulties are the large amount of training data, in order to assure the complete understanding of the attitude dynamics by the neural net, and also the fact that attitude control is a MIMO instead of a SISO system. Regardless of these draw backs, some results concerning attitude control will be shown