Aiming　at　the　problem　that　the　energy　consumption　and　time　indexes　are　affected　by　the　dynamic　mass　of　train　in　the　operation　of　automatic　train　operation　system　(ATO),　a　method　of　generating　time-energy　consumption　alternative　sets　for　subway　trains　based　on　the　spatio-temporal　data　of　passenger　flow　is　proposed.　Firstly,　this　paper　integrates　time　periods　and　stations　to　cluster　passenger　flows　using　K-means　algorithm　to　get　the　classification　of　passenger　mass　between　different　stations　under　different　time　periods,　and　calculates　the　approximate　mass　of　train　operation　based　on　the　classification,　replacing　the　commonly　used　fixed　mass　of　AW0-AW3　with　a　train　mass　that　is　closer　to　reality.　Then,　an　improved　adaptive　NSGA-II　algorithm　is　proposed　to　optimize　the　operating　curves　of　trains　with　dynamic　mass　to　obtain　the　Pareto　frontier　time-energy　consumption　alternative　set　of　time　and　energy　consumption,　which　can　be　further　used　for　the　optimization　of　multi-objective　train　schedules.