Real-time　and　accurate　short-term　traffic　flow　prediction　can　effectively　improve　the　efficiency　and　safety　of　the　transportation　system.　However,　complex　traffic　systems　are　highly　nonlinear　and　random,　which　makes　short-term　traffic　flow　prediction　a　challenging　issue.　In　recent　years,　deep-learning　based　methods　have　been　widely　applied　in　short-term　traffic　flow　prediction.　Particularly,　the　long　short-term　memory　neural　network　(LSTM)　model　bears　great　potential　for　its　capability　in　learning　from　temporal　information.　In　this　paper,　an　improved　LSTM　model　is　used　to　predict　the　short-term　traffic　flow　of　a　target　road　section　of　the　East　4th　Ring　Road　of　Beijing,　and　to　analyze　the　influence　of　different　input　configuration　on　prediction　accuracy　as　well.　Experimental　results　demonstrate　that　feeding　upstream　flow　and　velocity　information　does　improve　its　overall　performance.　Especially　after　traffic　flow　information　is　fed　with　corresponding　temporal　information,　the　accuracy　of　traffic　flow　prediction　has　been　significantly　improved.