In　intelligent　traffic　system,　short-term　traffic　flow　prediction　is　one　of　the　key　technologies　of　traffic　control　and　traffic　guidance.　Due　to　the　inaccuracy　of　Markov　traffic　flow　prediction　model,　according　to　the　characteristics　of　traffic　flow,　a　Markov　particle　filter　traffic　flow　prediction　model　is　proposed.　On　the　one　hand,　after　pretreatment　of　traffic　flow,　it　can　be　used　as　sample　data　to　predict　future　traffic　flow　in　Markov　model,　which　can　better　describe　the　trend　of　traffic　flow.　On　the　other　hand,　in　view　of　the　inaccuracy　of　the　prediction　results　and　the　disadvantages　of　non-linear　prediction　instability,　particle　filter　algorithm　is　used　to　update　the　prediction　results　and　weights,　and　the　sample　re-selection　process.　After　several　iterations,　the　sample　particles　are　closer　to　the　actual　prediction　result,　thus　improving　the　prediction　accuracy.　Finally,　the　traffic　flow　detected　by　a　detector　in　Changping　District　of　Beijing　is　simulated,　and　the　prediction　results　are　compared　with　the　traditional　Markov　chain.　The　results　show　that　the　5-minute　interval　error　and　1-hour　interval　error　of　the　proposed　Markov　particle　filter　traffic　flow　prediction　model　are　6.14%　and　6.04%　respectively.　It　shows　that　the　model　has　better　applicability　and　stability,　and　the　prediction　accuracy　is　high.　Copyright　©　2019　by　Science　Press.