【精选】一种导航卫星中长期轨道预报方法.pdfVIP

40 V ol. 40, Sup. 20 11 5 A cta Geodaet ica et Cart ogr aphi ca Sini ca M ay , 20 11 : 2011 S003907 1 1 2 1 周建华 , 杨 龙 , 徐 波 , 谢廷峰 1. , 100094; 2 . , 2 10016 A L ong Dur ation Orb it Pr ediction M ethod for Navig ation Satellit e 1 1 2 1 ZHO U J ianhua , Y A NG L ong , X U Bo , XIE T ing f eng 1. Beiji ng Global Inform at ion A ppl ication and D evelopment Cent er, Beij ing 100094, China ; 2. N anjing U nivers it y of A eronaut ics and A s t ronaut ics, N anjing 2 100 16, Ch ina Abstract : A hy brid pr edictio n m odel based o n neur al netw o r k is intr oduced fo r na vig ation satellite. T he new mo del is ba sed o n dy namica l mo del and the neural netw o rk mo del is a do pted to m odify it . Dur ing the tra ining phase, neural net w or k mo del tr ies to a ppr oach the diff er ence of dy na mical mo del pr edictio n pro duct , then com pensate it in t he pre dict ion phase. T he character istics o f D M M prediction err o r for G PS ar e ex plo r ed, and then pr editio n str ateg y fo r long term ( lo nge r than 15 d is desig ned . A t the last , a g ro up o f tests o n G PS satellit es ar e pro cessed to pr ov e and ev alu at e the pe rfo rm ance o f the new metho d , the test r esults sho w that the im pro vem ent fo r 15 ~ 40 d prediction is obv i o u s . Key w o rds : o rbit predictio n; artif icial neural netw o rk ; dynamical mo del ; hybr id mo del : 基于神经网络混合建模的思想提出一种针对导航 卫 的中长期轨道预报方法, 在原动力学模型的基础上引入神 经网络模型作为补偿, 从而获得新的预报模型在训练过程中神经网络通过学习动力学模型轨道预报误差来掌握其变 化规律, 并在预报过程中为动力学模型预报提供补偿, 以提高预报精度对GP S 卫 动力学模型中长期预报误差的特 点进行分析, 然后根据所得结论提出混合模型的中长期( 15 d 以上 预报方案, 最后通过对GPS 卫 的仿真试验证明混 合模型的改进效果, 结果表明新方法在 15~ 40 d 的预报上表现出很好的改进效果 : 轨道预报; 神经网络; 动力学模型; 混合建模 : P 228 : A