a pp licationofartificialneuralnetworktoorbitpredictionofbeidounavi g ationsatellites

autonomousorbitdeterminationofsatellitesisimp ortanttoimp rovetheavailabilit yofasatellitenavi g ations y stem,andde p endsonhi g h-p recisionorbitp redictions.theorbitp redictedwithad y namicsmodelhasthep roblemofhi g hdilution.tosolvethisp roblem，amethodisp ro p osedtoimp rovelon g-termorbitp redictionsforbeidousatellitesbasedonanartificialneuralnetwork（ ann）model.wedevelo p edanannmodelbasedonthed y namicsmodel ,inordertodeterminethevariationcharacteristicsintheorbitp redictionerrorsb ylearnin gandtrainin ghistoricalorbitp redictionerrors.weusedthisann modeltoimp rovetheaccurac yoforbitp redictionsb yestimatin gandcorrectin gp redictionerrors.formedium－termandlon g-termorbitp redictions,ex p erimentalresultsshowedthatorbitp redictionerrorsafterthea pp licationofthep ro p osedann modelwerelessthanthosebasedonthed y namicmodel.theeffectivenessoftheimp rovementsvarieswithdifferentsatellitesandinitiale p ochs.theerroroforbitp redictionsforthe15-da yp redictionwasreducedto19mfrom318m;and，forthe30-da yp rediction，wasreducedto49mfrom1757m.theimp rovementratiosforthe15-da yand30-da yp redictionswere41％～80％ and32％～88％,res p ectivel y.