A　new　algorithm　based　on　FCM,　Sub_FCM　is　presented.　This　algorithm　is　combined　with　subtractive　clustering　and　cluster　validity　evaluation.　It　can　be　used　to　determine　the　optimal　cluster　number　and　initialize　clustering　prototype　automatically.　An　automatic　classification　model　based　on　Alternative　Optimization　strategy　and　an　unsupervised　machine　learning　method　are　proposed.　Experiments　show　that　the　internal　structure　of　the　traffic　flow　data　is　revealed　correctly　and　the　proper　traffic　flow　range　is　determined　automatically.　The　results　lay　the　foundation　for　applying　a　proper　control　algorithm.