Expressway　plays　an　indispensable　role　in　the　transportation　of　goods　and　the　travel　of　residents.　accurate　traffic　flow　forecasting　is　of　great　significance　to　traffic　monitoring.　There　are　complex　spatiotemporal　correlations,　sequential　relations　and　heterogeneity　among　traffic　flows.　Aiming　at　the　problem　that　the　spatiotemporal　correlations　cannot　be　extracted　effectively,　an　optimization　model　for　traffic　flow　forecasting　is　proposed.　The　preprocessing　algorithms　such　as　data　cleaning　and　data　reconstruction　are　studied,　and　the　spatial　distribution　feature　and　periodic　variation　of　traffic　flow　in　ETC　data　are　analyzed.　Based　on　the　spatiotemporal　feature　of　highway　traffic　flow,　the　spatiotemporal　feature　vector　model　of　traffic　flow　is　constructed　to　extract　the　potential　spatiotemporal　feature　information　of　massive　ETC　data　quickly　and　effectively.　And　a　random　forest　algorithm　is　proposed　to　train　and　learn　features　of　data　to　predict　highway　traffic　flow.　It　is　verified　by　using　ETC　data　of　Expressway　in　Fujian　Province.　The　results　show　that　the　periodic　feature,　spatial　feature　and　adjacent　time　feature　in　the　spatiotemporal　feature　vector　model　effectively　improve　the　forecasting　accuracy.　The　results　show　that　the　SF-RF　model　has　better　prediction　accuracy　than　HA,　LR,　GBDT　and　other　traditional　models,　which　can　provide　reference　for　decision-making,　analysis　and　scheduling　of　intelligent　highway　management　system.　©　2021　ACM.