Research on Cooperative Orbit Determination in Satellite Network Based on Multi-Agent System Theory
|School||Harbin Institute of Technology|
|Course||Information and Communication Engineering|
|Keywords||satellite network moving dynamic coalition reinforcement learning of case cooperative orbit determination particle filter|
Satellite network embodies scout satellites, guarding satellites, communication satellites and navigation satellites, which is running autonomously and with the capability of obtaining, fusing, storing and distributing the information independently. So it has become an important armament for capturing the information predominance. How to introduce the intelligence technology into the application of satellite network, such as fuzzy logic, neural network, reinforcement learning and multi-agent system, is an important aspect in the improvement of sensor network. That can enhance the autonomy, cooperation and recovery ability of satellite network.Under the background of manoeuvrable space targets tracking in satellite network, the main work of this thesis is to explore the appropriate scheme about intelligence information processing and cooperation technology at the sensor network field. A reinforcement learning scheme of case, based on fuzzy neural network and Q-learning, and EKF-based particle filter was proposed. The problems of moving dynamic coalition learning and manoeuvrable space target tracking are solved.In the study of intelligent cooperation technique in sensor network, we use a new scheme for the formation of moving dynamic coalition, which integrates case-based reasoning , neural network and reinforcement learning. A reinforcement learning scheme of case, based on fuzzy neural network and Q-learning, was proposed to solve the learning problem of moving dynamic coalition. The scheme improved the coalition quality and formation efficiency. The self-learning in coalition formation is also realized in dynamic networks. Target tracking is a typical state-estimation problem and the state equation of space targets is strong non-linear. The key to solve this problem is to find a filter arithmetic. The state and observation model was constructed based on the dynamics analytic of satellite. The basic theory and disadvantages of particle filter are presented. Then the improved particle filter based on extended Kalman filter is designed to track satellite with the ability of orbit transfer. Simulation results show that this method is preponderant and produces smaller error compared with classical method.In addition, due to the characteristic of satellite network, we discussed the formation and learning of moving dynamic coalition in satellite network, based on the analysis of the condition and principle of satellite tracking particularly. The intelligent cooperation technique in satellite network is also used to track manoeuvrable space targets. Simulation results show that the learning scheme based on FNN-Q ameliorated the formation of satellite coalition and reduced the error of tracking. And the performance of surveillance and tracking is improved a lot in satellite network.