Prediction Model for Traffic Flow with Missing Values Based on Generative Adversarial and Graph Convolutional Networks

Abstract

In order to improve the accuracy of urban road network traffic-flow prediction with missing values, the generator and discriminator of the generative adversarial network were reconstructed, the loss function was improved, and the traffic generative adversarial imputation network (TGAIN) was proposed for the completion of the missing data in traffic flow. Based on empirical mode decomposition (EMD), graph convolutional networks (GCN), and gated recurrent unit (GRU), the EMD-GCN-GRU model was designed for urban road network traffic-flow prediction. First, the traffic-flow data was processed using empirical mode decomposition and each component of the same level was reconstructed as the input of the subsequent prediction model. Then, the graph convolutional networks were used to learn the road network topology to capture the spatial characteristics of the traffic flow, and the gated recurrent unit was employed to capture the temporal characteristic of traffic flow. For the road network traffic-flow data with missing values, TGAIN was used to complete the data, and then the EMD-GCN-GRU was used to predict the traffic flow. The Shenzhen average vehicle speed data set was used to construct a variety of typical traffic-flow data with different missing patterns and different missing rates to simulate the actual missing situation. The effectiveness of the method was verified on the ModelArts development platform. The results show that compared with the commonly used matrix factorization imputation method, the TGAIN model has higher completion accuracy in the random missing mode of the data set and has better completion performance when the nonrandom missing rate is lower than 50%. Compared with seven other prediction algorithms, the proposed prediction method has higher prediction accuracy. Combining the data imputation method TGAIN with the traffic-flow prediction method EMD-GCN-GRU for urban road network traffic-flow prediction with missing values can significantly reduce the negative impact of missing data and data noise on traffic-flow prediction and capture the spatial and temporal correlation of network traffic flow, which improves the accuracy of urban road network traffic-flow prediction.