Reasonable　planning　of　passenger　evacuation　routes　can　directly　improve　evacuation　efficiency　in　the　metro　station.　Considering　the　effect　of　traffic　capacity　at　the　node　such　as　gate　and　stair/escalator,　a　multi-objective　optimization　model　aimed　at　minimizing　the　maximum　evacuation　time　of　heterogeneous　passengers　in　different　segmentation　areas　is　established,　which　achieves　local　optimization　on　the　basis　of　global　optimization　of　the　evacuation　route.　Node　traffic　capacity　is　assessed　based　on　the　combination　of　BP　neural　network　and　atomic　orbital　search　algorithm,　which　is　conducive　to　improving　the　calculation　accuracy　of　total　evacuation　time.　The　simulation　results　indicate　that　the　performance　of　the　established　prediction　model　of　node　travel　time　is　good,　and　the　mean　square　error　can　be　as　low　as　0.2555.　Meanwhile,　the　proposed　route　strategy　can　improve　the　overall　evacuation　efficiency　by　21.85%,　and　passengers＇　choices　of　evacuation　routes　are　more　reasonable.　The　sensitivity　analysis　of　factors　affecting　evacuation　through　a　random　forest　algorithm　shows　that　the　number　of　passengers　with　person　attribute　E　has　the　greatest　impact　on　evacuation　time,　which　should　be　the　focus　of　attention　in　the　evacuation　process.