<正>Someproblemsinusingv-supportvectormachine(v-SVM)forthepredictionofnonlineartimeseriesarediscussed.Theproblemsincludeselectionofvariousnetparameters,whichaffecttheperformanceofprediction,mixtureofkernels,anddecompositioncooperationlinearprogrammingv-SVMregression,whichresultinimprovementsofthealgorithm.ComputersimulationsinthepredictionofnonlineartimeseriesproducedbyMackey-GlassequationandLorenzequationprovidesomeimprovedresults.