Three-axis Stabilized Geostationary Satellite Attitude Control Using Neural Predictive Algorithms

Abstract – A study of a three-axis stabilized satellite attitude control in geostationary orbit is presented. The analysis is based on a neural predictive approach using Kalman filtering algorithms. Optimization of a quadratic performance functional is used to determine the discrete predictive control actions. The control determination on a typical iteration is viewed as a stochastic optimal linear parameter estimation problem. Direct analogy with Kalman filtering algorithms allows the derivation of full non-parallel as well as approximated parallel processing algorithms. The satellite control system is based on a gyro device, which furnishes control torques on all three vehicle’s axes. Therefore, wheel speed control plus two-degree-of-freedom gyrotorquing supply the required moments to counterbalance attitude perturbations due to the solar pressure torques, limited orbit control thruster misalignment, or small initial satellite attitude misalignment. The results demonstrate that the proposed neural predictive scheme, at only one step-ahead prediction furnishes smooth control actions required to point and to maintain the satellite stabilized in the desired direction. Simulations are presented for a one-day satellite attitude control and for small initial attitude angles misalignment.