Urban traffic flow prediction is an important part of building a green, low-carbon, safe and efficient intelligent transportation system. Spatiotemporal graph neural networks are widely used in urban traffic flow prediction due to their strong spatiotemporal data representation capabilities. However, most of the current models have a series of problems such as insufficient spatial information mining of traffic flow data, and based on this, a traffic flow prediction model based on convolutional neural network of time-domain graph is proposed. With the help of dilated causal convolution, the perceptual field is expanded, and the time information is extracted by combining with the residual network. Finally, the specific dataset is used for evaluation, and the results show that the model has better performance than the graph convolutional network model and the multi-spatiotemporal graph convolutional network model.