In　order　to　improve　the　prediction　accuracy　of　the　atmospheric　pollutants　concentration,　the　gray　correlation　analysis　is　used　to　select　the　main　factors　affecting　the　PM2.5　concentration　in　the　atmosphere.　Regarding　them　as　the　input　variables,　a　model　based　on　BP_Adaboost　neural　network　is　proposed　to　predict　the　PM2.5　concentration.　The　modified　particle　swarm　optimization　(MPSO)　algorithm　is　applied　to　choose　the　weight　and　threshold　of　BP_Adaboost　neural　network,　which　can　availably　avoid　falling　into　local　optimal　solution.　According　to　the　concentration　of　air　pollutants　and　meteorological　condition,　the　data　between　November　1,　2014　to　November　25,　2014　and　July　7,　2017　to　August　6,　2017,　which　are　monitored　every　hour　by　the　Wanliu　station　of　Haidian　distinct　and　Beijing　University　of　Technology　point　of　Chaoyang　distinct　in　Beijing,　are　used　as　the　experiment　object.　The　simulation　results　show　that　the　PM2.5　concentration　prediction　performance　of　MPSO-BP_Adaboost　neural　network　is　better　than　that　of　BP_Adaboost,　BP　and　generalized　regression　neural　network.　©　2018,　Editorial　Office　of　Journal　of　Dalian　University　of　Technology.　All　right　reserved.