Formation control in low-Earth orbits by means of machine learning methods

In this paper, we design an adaptive neurocontroller for keeping a formation of two spacecraft in low-Earth near-circular orbits. One of the spacecraft is controllable, the other is not. The controllable spacecraft is able to change the cross section within known limits, as well as apply velocity impulses in any direction. Main approaches to solve the problem are discussed; two control neural networks are introduced, and their optimal size is found. The problem of adaptation of two uncertain parameters — the ballistic coefficient of the uncontrolled spacecraft and the atmospheric density — is also considered. The adaptation problem is solved by a non-gradient optimization method.