Abstract: InordertogetrealtimeGPSclockproductsofhighaccuracy,anewpredictionalgorithmwasproposed,whichincorporatedapolynomialmodelwithafewcyclictermsandleastsquaressupportvectormachines(LSSVM)forclockoffset.Firstly,apolynomialmodeladdingafewcyclictermswasutilizedtofittheclockoffsetseriesaccordingtothephysicalcharacteristicsofsatelliteclockssoastoextractthetrendtermandperiodicterm.Then,thepolynomialfittingresidualsweremodeledbasedontheLSSVM.Finally,thetrendtermandperiodictermwereaddedtotheforecastedresidualresulttoproducethefinalpredictionvaluefortheclockoffset.TheIGSultrarapidobserved(IGUO)productswereemployedastheinitialvaluesofthepredictionstrategy.Thesimulationresultsdemonstratethattheproposedalgorithmcanbeusedtopredictclockoffsetwithhighaccuracyinrealtime,andoutperformstheIGSultrarapidpredicted(IGUP)solutionsatleastonadailybasis.

Key words: Leastsquaressupportvectormachines, Ultrarapidephemeris, Predictionmodel, GPSsatelliteclocks