基于神經網絡的混合模型軌道預報方法研究.pdf

1、南京航空航天大學碩士學位論文基于神經網絡的混合模型軌道預報方法研究姓名：董澤政申請學位級別：碩士專業(yè)：導航制導與控制指導教師：徐波2010-12基于神經網絡的混合模型軌道預報方法研究 II Abstract Satellite orbit predic

2、tion is a process to calculate the motion state in certain duration, it is essential for orbit design, satellite tracking and GPS positioning. The traditional method for prediction is based on Newton second law, which ne

3、eds highly precise dynamical models. As the space environment is complex and dynamic variation, it is doomed to take a long time and much works to establish the dynamical models of satellite and even more time and works

4、to consummate it. So the accurancy of orbit perdiction based on dynamical models is very hard to improved. In this paper a method of satellite orbit forecasting based on Artificial Neural Network (ANN) is proposed, the h

5、ybrid prediction model consists of ANN and dynamical models (DMM). In order to acquire more highly precise ephemeris, during the training phase, ANN tries to approach the difference between the IGS ephemeris and the DMM

6、predicting product.. The main study in this paper are followed: 1.The highly accurancy tansition process between the GCRS(Geocentric Celestial Reference System) and the ITRS(International Terrestrial Reference System), i

7、ncluding the Precession-nutation model, the Earth-Rotation model and the EOP( Earth Orientation Parameters); 2.The strategy of orbit predition with high precision, consisted of analysis for dynamical models and research

8、on the numerical integrators; 3.The strategy of orbit determination, preliminary orbit determination and parameters estimation algrithms are needed in this part; 4.The training algrithm and method of improving generaliza

9、tion ability; 5.Hybrid model sysytem design and the predtion strategies. We choose GPS satellites as the researching object, first we carry out the predition of GPS satellites, and then the short duration and long durati

10、on characteristics of prediction errors were explored, and based on these, we obtained two predtion strategies for the Hyrbrid Model The test results shows that the 24h prediction error of Hyrbrid Model is less than 1 me