Prediction of traffic flow based on deep learning

Deep neural networks (DNNs) have recently demonstrated the capability to predict traffic flow with big data. Although existing DNN models can provide better performance than shallow models, it is still an open question to make full use of the spatio-temporal characteristics of traffic flows to improve performance. We propose a novel deep architecture combining CNN and LSTM for traffic flow (RCF) predictio. The model uses CNN to explore temporal correlation and LSTM to explore spatial correlation . Factors such as weather and historical period data are also added to the feature. Its advantage lies in making full use of the spatial-temporal correlation of traffic data and more comprehensively considered the impact of multiple related factors. Aiming at the difficult problem of obtaining spatial features, a feature selection method based on Random Forests is proposed. We use the gini score to represent the spatial connection between intersections to form a network graph constructed based on data.  The experimental results show that based on the random forest feature selection and RCF model, the accuracy of traffic prediction reaches 90%