Abstract: With the continuous advancement of smart city construction, urban short-term traffic flow forecasting becomes more and more important. According to the influence of traffic flow characteristics and external factors on traffic flow forecast results, the model CNN-ResNet-LSTM for urban short-term traffic flow forecasting is proposed. The model integrates convolutional neural networks(CNN), residual neural units(ResNet) and long-short-term memory networks(LSTM) into an end-to-end network framework. The convolutional neural network is used to capture the local spatial characteristics of traffic flow, and multiple residual neural units are added to deepen the network depth and improve the prediction accuracy. On the other hand, the long short-term memory-cycle neural network is used to capture temporal characteristics of traffic flow data. The output results of the two networks are combined by the corresponding weights to obtain the predicted results through the trajectory data. Finally, the traffic flow prediction values of the urban areas are obtained by fusing with external factors. Through the verification of the CNN-ResNet-LSTM model by Beijing data, the model is not only higher in accuracy than the traditional model, but also has fewer parameters in the case of ensuring the accuracy of prediction.

Key words: traffic flow forecasting, deep learning, long short-term memory network, convolutional neural network