Support　Vector　Machine　is　applied　to　the　modeling　of　a　nonlinear　dynamic　system.　Linear　kernel　is　adopted　in　sample　training　and　the　parameters　in　the　mathematical　model　are　calculated　by　resultant　lagrangian　factors　and　support　vectors.　To　diminish　the　parameter　drift　in　identification,　training　samples　are　reconstructed　by　difference　method.　Correlation　analysis　demonstrates　the　validity　of　reconstruction.　Based　on　the　regressive　mathematical　model,　the　dynamics　of　the　system　is　predicted　and　comparison　between　predicted　results　and　test　results　confirms　the　parameters　identified.　©　2012　IEEE.