A Survey of Short-Term Traffic Volume Prediction Methods Based on Composite Models

Abstract

Traditional traffic prediction methods cannot effectively use many traffic data. Deep learning can mine the information behind big data. For example, recurrent neural network can effectively extract time rules. Convolution neural network can extract spatial features. And, graph convolution neural network is convenient for graph data processing, but it still has its limitations. At present, the hybrid method highlights its advantages in the field of transportation and realizes the complementary advantages of traditional methods and deep learning methods. On this basis, this paper summarizes the short-term traffic volume forecasting methods.