Comprehending　the　variation　in　traffic　flow　is　critical　to　alleviating　traffic　congestion　at　expressway　toll　station　exits.　Despite　the　fact　that　various　traffic　flow　forecasting　models　have　been　proposed,　most　of　them　make　predictions　based　on　the　entry　traffic　in　the　area　near　the　target　toll　station.　For　origin–destination　data　like　from　entry　to　exit,　these　methods　can　hardly　capture　information　on　vehicles　in　transit.　In　this　work,　we　suggest　for　the　first　time　predicting　toll　station　exit　flows　based　on　expressway　gantry　data.　Moreover,　in　order　to　obtain　the　contribution　of　multiple　gantry　series　to　the　exit　traffic　flow,　a　recurrent　neural　network　incorporating　spatio-temporal　attention　mechanism　is　proposed.　The　proposal　not　only　predicts　effectively　but　also　improves　the　interpretability　of　the　model.　Comparative　experiments　were　conducted　using　data　from　the　gantry　system　and　toll　station　data　of　the　expressway　in　Fujian　Province,　China.　The　experimental　results　show　that　the　proposed　model　performs　better　than　other　baseline　methods.　©　2023,　The　Author(s),　under　exclusive　license　to　Springer　Nature　Singapore　Pte　Ltd.