Accurate　and　stable　short-term　passenger　flow　prediction　is　an　indispensable　part　of　current　intelligent　transportation　systems.　This　paper　proposes　two　deep　learning　prediction　models　based　on　convolutional　neural　networks　(CNN)　and　long　short-term　memory　neural　network　(LSTM).　Combining　the　CNN　characteristics　and　the　LSTM,　the　ConvXD-LSTM　extracts　passenger　flow　features　through　CNN　and　then　inputs　the　time　series　into　the　LSTM.　The　ConvLSTM　converts　the　weight　calculation　of　the　LSTM　into　convolution　operation　to　realize　short-term　passenger　flow　prediction.　Fuzhou　Metro　Line　1　passenger　flow　data　was　used　for　verification.　The　models　were　used　to　predict　the　passenger　flow　of　subway　stations　and　cross-sections　and　compared　with　the　traditional　prediction　models.　In　terms　of　prediction　accuracy,　ConvLSTM　has　the　highest　accuracy,　followed　by　ConvXD-LSTM.　In　terms　of　running　time,　ConvXD　is　the　fastest　and　LSTM　takes　the　longest.　ConvXD-LSTM　and　ConvLSTM　are　in　the　middle　of　the　two　models,　achieving　a　good　balance　between　accuracy　and　efficiency.　Compared　with　ConvXD-LSTM,　ConvLSTM　has　a　relatively　simple　network　structure,　which　reduces　the　computational　burden　and　improves　the　prediction　accuracy.