With　the　rapid　development　of　society,　the　number　of　vehicles　is　increasing　rapidly,　Service　groups　for　expressway　are　increasing.In　order　to　improve　the　operation　and　management　effect　of　expressway　and　alleviate　the　congestion　phenomenon　caused　by　excessive　vehicles,　the　precise　short-term　traffic　flow　prediction　of　expressway　has　become　an　important　research　topic.　In　this　paper,　a　combined　model　method　is　proposed　to　predict　the　short-term　traffic　flow　of　expressway.　The　method　utilizes　RNN　capture　different　lag　features　in　time　series　and　attentional　mechanism　to　link　the　acquired　lag　features　and　historical　data　to　Seq2seq　model　to　predict　complex　nonlinear　traffic　flow　data.　At　the　same　time,　we　interpolate　the　missing　value　of　the　data.　Compared　with　the　original　model　Seq2seq,　the　precision　of　combined　model　is　improved　by　67.0%.　Compared　with　GRU,　LSTM　and　other　traditional　single　model,　the　combined　model　has　better　precision.　©　2021　ACM.