Urban road travel time prediction based on gated recurrent unit using internet data

Abstract

Traffic congestion has been aggravated in many cities, which not only negatively affects the traffic efficiency, but also exacerbates the related social problems. Accurate prediction of urban road travel times plays a crucial role in assisting traffic management and alleviating the derived problems caused by traffic congestion. This study aims to propose an innovative Gated Recurrent Unit (GRU)-based model for vehicle travel time prediction on urban road networks using the Internet vehicle travel time data. The Internet data used in the paper represents the real-world average vehicle travel times on urban road networks, avoiding the privacy issue and the computational challenge in individual trajectory data. Before presenting the model, a data imputation method based on time series to reconstruct the missing data, and a time-series data similarity evaluation method for road link classification are developed. The road-category-based model helps balance the computational efficiency and the prediction accuracy. Finally, the modelling framework is tested on a road network in Xuhui District, Shanghai. By comparing its performance with different models, it is concluded that the GRU-based model considering road link categories is more efficient and more accurate in urban road travel time prediction.