Deep learning based velocity prediction with consideration of road structure

Speed prediction of roads is an important part of intelligent transportation system(ITS). Due to the complexity of the traffic environment, the speed of road is not only related to the historical speed of this road, but also has a strong relationship with the adjacent roads. In recent years, with the development and application of various data acquisition devices, a large amount of real-time traffic status information can be obtained. In this paper, we propose a road speed prediction method based on deep learning. Firstly, the method extracts the spatial correlation between the target road and the adjacent roads by convolutional neural networks(CNN). Secondly, long short-term memory(LSTM) is used to obtain the temporal correlation of the data and effectively capture the nonlinear features in the traffic speed sequence. Finally, a multilayer perceptron(MLP) is used to combine the spatio-temporal features. In this paper, the prediction accuracy of the model is evaluated on a real traffic dataset. Compared with traditional statistical methods and advanced deep learning methods, the speed prediction model in this paper obtains a more accurate prediction performance.