In order to tackle existing traffic flow prediction problem, a Traffic Volume Forecast Model based on deep learning is designed. The model implements Convolutional Neural Network (CNN) to extract spatial matrix information, uses long and short-term neural network (LSTM) for sequence prediction, appends attention mechanism to time step on LSTM, and assigns weights to different time steps. By implementing model verification on the Chengdu taxi dataset, dividing data into various categories, cross validating different categories of data, and comparing the model with other models, it is concluded that the CNN-LSTM-At network model proposed in this article has higher accuracy compared with traditional network model.