The invention fully utilizes space-time characteristics of urban population flow data, and on the basis of residual error network structures ResNet and DenseNet, provides an urban population flow prediction method based on a double-path space-time residual error network. The method adopts the double-path space-time residual error network to perform urban population flow prediction, and can improvearea population flow-in and flow-out prediction performance. The double-path space-time residual error network includes a plurality of study modules; each study module consists of micro blocks, eachmicro block starts from a first convolutional layer, is then connected with a second convolutional layer, and ends with a third convolutional layer; and outputs of the third convolutional layer include a first output and a second output, the first output is added to an original signal-channel residual error network according to a corresponding element adding method, and the second output is connected with a path of a dense connection network. The urban population flow prediction method based on a double-path space-time residual error network utilizes the space-time characteristics of urban population flow data, and on the basis of the residual error network structures ResNet and DenseNet, designs a double-path residual error network ST-DPResNet, thereby further improving the area population flow-in and flow-out prediction performance.