The invention discloses a regional network flow prediction method based on deep learning, and the method comprises the steps: 1, obtaining a regional network flow sequence, and carrying out the statistics of a flow value of the regional network flow sequence at each moment; 2, extracting a flow matrix sequence with a corresponding characteristic as the input of a deep learning prediction model according to the spatial correlation and time correlation of the regional flow sequence, wherein the time correlation comprises compactness, periodicity and tendency; 3, for the three input flow matrix sequences obtained in the step 2, extracting time and space correlation by using a 3D convolutional neural network and ConvLSTM respectively; and 4, fusing the features, extracted from 3D convolution and ConvLSTM, of the three flow matrix sequences, and carrying out the final flow prediction based on an attention mechanism. According to the method, the periodic change characteristic of the flow sequence is covered under the limited input length through the time sequence extraction method, the regional network flow value at the next moment is predicted with high accuracy, reasonable distributionof wireless resources is facilitated, and the resource utilization rate is increased.