The　improvement　of　traffic　flow　prediction　accuracy　is　of　great　significance　for　an　Intelligent　Transportation　System.　However,　most　current　prediction　methods　are　based　on　the　complete　or　relatively　complete　datasets.　However,　complete　traffic　datasets　are　not　easy　to　obtain.　In　this　paper,　we　propose　a　3D　Convolutional　Ambient　Generative　Adversarial　Network　to　predict　traffic　flow　by　using　the　incomplete　datasets.　The　proposed　model　is　able　to　learn　the　underlying　distribution　of　traffic　flow　from　the　incomplete　traffic　data　and　utilize　the　captured　spatio-temporal　features　of　traffic　data　for　traffic　flow　prediction.　In　addition,　we　also　introduce　an　attention　mechanism　into　the　model　to　improve　the　prediction　accuracy　by　exploring　the　global　regional　structure　correlation.　The　simulation　results　demonstrate　that　the　proposed　model　outperforms　the　state-of-the-art　prediction　methods　for　traffic　flow　prediction　when　incomplete　data　is　utilized　for　training.　©　2019　Association　for　Computing　Machinery.