Intersections　are　the　key　node　in　the　urban　road　network,　so　reasonable　channelization　at　intersections　is　key　for　improving　traffic　efficiency　of　the　entire　urban　network.　However,　traffic　data　of　intersections　that　has　been　collected　so　far　has　great　volatility　and　abnormality,　which　cannot　provide　an　accurate　data　basis　for　further　intersection　optimization.　This　paper　is　based　on　historical　traffic　data　of　intersections　for　data　processing　and　short-term　traffic　forecasting.　First,　the　historical　data　is　preprocessed　by　a　time　series　method　and　short-term　traffic　prediction　method　to　recover　the　missing　data.　We　then　performed　short-term　traffic　forecasting　based　on　SPSS　and　used　an　expert　modeling　method　and　ARIMA　forecasting　method　to　predict　short-term　traffic.　After　pretreatment,　we　performed　time　division　of　traffic　data　using　the　K-means　clustering　algorithm.　Through　the　above　methods,　traffic　data　can　be　improved　to　provide　accurate　data　support　for　intersection　optimization.　©　ASCE.