SAE Network: A Deep Learning Method for Traffic Flow Prediction

Traffic flow forecasting is one of the most important work of the intelligent transport system (ITS). It has extensive and promising applications to improve the traffic conditions. However, the forecasting accuracy may not be high enough for traffic control considering high-dynamic change and various disturbance in the real world. In order to improve the forecasting accuracy, huge amount of historical traffic data are studied. In this paper, four months of aggregated traffic flow data via vehicle detector on a city ring way are obtained and selected for further analysis. An improved stacked autoencoder (SAE)model based on deep learning network is proposed to extract the features among traffic flow data. This predict model is trained in a greedy layer-wise method. The prediction results illustrate that this proposed model with higher accuracy is superior to other methods for traffic flow prediction.