Artificial　neural　network　has　been　widely　used　in　displacement　back　analysis,　but　it　has　the　problems　of　large　sample,　over-fitting,　local　optimization　and　poor　generalization　performance,　so　it　has　the　poor　adaptability　in　the　Geotechnical　Engineering.　Support　Vector　Machines　algorithm　has　the　advantages　of　small　sample,　global　optimization　and　generalization　performance.　A　direct　optimization　method　based　on　genetic　algorithm　and　the　improved　support　vector　regression　algorithm　(GA-SVR)　is　applied　in　order　to　identify　multinomial　parameters　intelligently　and　forecast　displacements　fast　and　exactly,　combined　with　an　unsymmetrical　pressure　tunnel　with　shallow　depth　section　of　the　left　line　of　import　in　BEIKOU　Tunnel　on　Zhangjiakou-Shijiazhuang　highway.　The　application　result　shows　the　new　type　of　intelligent　displacement　back　analysis　could　obtain　accurately　the　parameters　of　rock　mechanics　and　initial　stress　in　limited　monitoring　data　and　provide　parameters　for　ahead-forecast　of　rock　deformation.