Abstract
Road traffic prediction is a vital role of real-time traffic management in the intelligent transportation system (ITS). Many of existing prediction models are established and achieved good results. However, most of them ignored the intrinsic characteristics of traffic parameters data and also not considered on the spatiotemporal effects of the road section which can reflect the situation of the whole road section traffic. Therefore, multi-node road section traffic prediction is still an open problem. This paper, the empirical mode decomposition (EMD) technique is employed to decompose the traffic parameters information into many intrinsic mode function (IMF) components, which represent the original road traffic information in periodic sequence and random sequence. Then, by considering the superiority of convolution neural network (CNN) in multi-dimensional data processing which could handle the spatiotemporal effects, a prediction model based on CNN is used to achieve the prediction of periodic sequence and random sequence. Finally, two parts of the prediction results are combined to get the final prediction results. The dataset from Caltrans Performance Measurement System is used for building the model and compared with several well-known models, such as PCA-BP, Lasso-BP, and standard CNN. The results show that the proposed prediction model achieves higher accuracy with smaller prediction error.