The　evacuation　process　after　the　earthquake　is　affected　by　different　shelters,　which　can　easily　lead　to　disorderly　evacuation　and　prolong　the　overall　evacuation　time.　Therefore,　the　scientific　planning　and　design　of　the　service　space　of　shelters　in　urban　disaster　prevention　areas　is　necessary　to　ensure　efficient　and　rational　evacuation.　In　this　paper,　Set　Pair　Analysis-Variable　Fuzzy　Recognition　Model　(SPA-VFRM)　was　used　to　analyze　the　risk　of　evacuation　disorder　in　urban　disaster　prevention　areas.　Besides,　the　method　for　dividing　the　service　scope　of　medium-　and　long-term　shelters　(MLS)　based　on　multi-agent　simulation　was　proposed　to　improve　the　evacuation　efficiency,　which　considered　the　characteristics　of　evacuation　space　and　the　people　composition　in　different　types　of　areas.　Taking　Xuchang　City　in　China　as　an　example,　the　multi-agent　dynamic　simulation　experiment　was　carried　out　based　on　the　evacuation　demand　after　the　earthquake.　The　results　showed　that　the　disorderly　evacuation　harmed　the　overall　evacuation　time　and　people　refuge　distribution　in　the　area.　When　the　service　radius　of　MLS　in　industrial,　residential,　and　comprehensive　areas　were　2.0　km,　1.0　km,　and　1.5　km,　respectively,　the　people　evacuation　in　the　area　could　be　completed　within　1　h,　which　met　the　requirements　of　the　long-distance　evacuation　(LDE)　time　in　the　specification.　The　research　results　can　provide　a　basis　for　the　quantitative　optimization　of　the　service　space　for　MLS　in　different　types　of　urban　areas,　and　enrich　and　deepen　the　design　theory　of　urban　disaster　prevention　shelters.