With　the　rapid　development　of　urban　rail　transit,　train　parking　accuracy　has　received　much　attention,　especially　for　subway　lines　with　platform　screen　doors.　In　actual　operation,　several　balises　are　mounted　on　the　track　to　enhance　the　parking　accuracy　by　providing　exact　positioning　data　for　the　train.　Currently,　the　number　and　positions　of　the　balises　are　determined　by　experience　and　iterative　experiments　that　may　greatly　increase　the　costs.　Combining　expert　knowledge　and　train　dynamics,　this　paper　formulates　a　balise　arrangement　optimization　(BAO)　model　to　study　the　relationship　between　the　number　&　locations　of　balises　and　parking　errors.　The　resistances,　nonlinearity　and　time　delay　in　train　braking　system　and　variable　initial　speeds　that　a　train　enters　the　parking　area　are　considered　in　the　formulation　of　BAO　model.　Moreover,　we　propose　a　genetic　algorithm　(GA)　to　solve　the　BAO　model　and　present　numerical　experiments　based　on　field　data　collected　from　Beijing　Subway　Yizhuang　Line.　The　results　indicate　that　the　BAO　model　can　enhance　the　parking　accuracy　to　about　0.10 m　via　changing　the　positions　of　the　balises.　Furthermore,　we　found　that:　(1)　more　balises　lead　to　better　performance　of　train　parking　accuracy;　(2)　four　or　five　balises　are　appropriate　for　balancing　the　device　cost　and　parking　errors.　©　2016　Elsevier　Inc.