Based　on　wavelet　decomposition　and　phase　space　reconstruction,　proposed　a　multi-resolution　SVR　traffic　flow　prediction　model.　The　complex　traffic　flow　sequence　is　transformed　into　a　single　character　at　different　scales　by　wavelet　transform,　and　the　sub-sequences　of　different　scales　are　predicted　by　the　SVR　model　respectively.　To　extract　more　key　information　from　the　time　series,　according　to　the　chaotic　characteristics　of　traffic　flow,　phase　space　decomposition　is　performed　to　improve　prediction　accuracy.　Experiments　show　that　the　proposed　method　can　better　reflect　the　sequence　details　and　variation　characteristics　and　achieve　higher　prediction　accuracy.