Stairs　are　the　bottleneck　areas　in　the　process　of　passenger　evacuation　in　the　subway　station.　The　safety　assessment　of　passengers　passing　through　the　stairs　helps　to　formulate　the　evacuation　plan　in　advance.　Firstly,　aiming　at　the　difficulty　of　collecting　the　evacuation　data　of　passengers　ascending　the　stairs,　MassMotion　simulation　software　was　adopted　to　build　a　stair　scene　to　simulate　the　evacuation　behavior　of　passengers　ascending　the　stairs,　and　the　basic　data　of　evacuation　time　were　obtained.　Then,　the　random　forest　model　was　trained　and　tested　with　basic　data　to　predict　the　evacuation　time　of　passengers　ascending　the　stairs.　Finally,　a　comprehensive　evaluation　model　of　evacuation　safety　was　established,　and　the　evacuation　safety　level　of　passengers　ascending　the　stairs　in　the　subway　station　was　evaluated　with　evacuation　time,　passenger　density　and　evacuation　panic　as　indicators.　The　research　results　indicate　that　mean　absolute　error(MAE)　of　the　prediction　results　of　the　random　forest　model　used　in　this　paper　is　3.　45　s,　and　mean　absolute　percentage　error　(MAPE)　is　3.　8%.　Compared　with　back　propagation　neural　network　(BPNN)　model　and　support　vector　regression　(SVR)　model,　the　prediction　accuracy　is　higher.　The　comprehensive　evaluation　model　of　evacuation　safety　is　used　to　evaluate　the　safety　of　the　stairs　in　a　subway　station　in　Qingdao,　and　the　evaluation　value　of　evacuation　safety　in　the　early　peak　period　is　medium.　©　2023　Research　Progress　of　Solid　State　Electronics.　All　rights　reserved.