Design Ofadaptive Neural Network Controllers For Drag-free Control System

To explore the astrodynamics, detect the earth’s gravitational waves and determinethe earth’s gravitational field, the satellite must follow a low earth orbit precisely, which isnamed as Drag-free satellite. The purpose of drag-free control system is to compensatenon-gravitational force, as well as the disturbance force (include atmospheric drag, solarradiation pressure and the ground reflection, etc.) through thrust, which ensure that thesatellite follow the orbit absolutely. This paper will investigate the uncertainty modelingand controller design for the drag-free control system of a low-orbital satellite with singleproof mass.Most researches have proposed controllers with linearized model and ignoring thenonlinear characteristics, which lower the accuracy of controllers. This paper will analyzethe nonlinear model of drag-free control system directly. First of all, the dynamics ofdrag-free satellite are modeled. For the nonlinear characteristics and unmodeled dynamics,RBF neural network is employed for approximation. In the case of ignoring cross coupling,the update laws of adaptive neural network weights is established and an adaptive neuralnetwork controller is designed based on Lyapunov methods and adaptive backsteppingcontrol theory, which guarantee the stability of the closed-loop system and satisfyrequirements of the drag-free satellite control system.Then, in consideration of cross coupling among subsystems, drag-free control systemcan be seen as interconnected system composing of three subsystems. The system isdecomposed into three subsystems,and then the controller is designed for each subsystemin isolation based on decentralized control strategy. Each controller only contains inputand output information of the subsystem. Finally, The simulation results indicate that thecontroller is effective.