STGNN-TTE: Travel time estimation via spatial–temporal graph neural network

Abstract

Estimating the travel time of urban trajectories is a basic but challenging task in many intelligent transportation systems, which is the foundation of route planning and traffic control. The difficulty of travel time estimation is the impact of entangled spatial and temporal dynamics on real-time traffic conditions. However, most existing works does not fully exploit structured spatial information and temporal dynamics, resulting in low accuracy travel time estimation. To address the problem,we propose a novel spatial–temporal graph neural network framework, namely STGNN-TTE, for travel time estimation. Specifically, we adopt a spatial–temporal module to capture the real-time traffic conditions and a transformer layer to estimate the links’ travel time and the total routes’ travel time synchronously. In the spatial–temporal module, we present a multi-scale deep spatial–temporal graph convolutional network to capture the structured spatial–temporal dynamics. Also, in order to enhance the individual representation of each link, we adopt another transformer layer to extract the individualized long-term temporal dynamics. Finally, these two parts are integrated by a gating fusion module as the real-time traffic condition representation. We evaluate our model by sufficient experiments on three real-world trajectory datasets, and the experimental results demonstrate that our model is significantly superior to several existing methods.