Travel time prediction using gated recurrent unit and spatio-temporal algorithm

Abstract

This paper proposes a method to predict highway travel times using a spatio-temporal algorithm based on a gated recurrent unit (GRU). The spatio-temporal algorithm predicts the travel times considering both spatial and temporal characteristics of travel time. It could reduce the time-lag problems between the experienced and predicted travel times on travel routes. The results of the spatio-temporal GRU were compared to those of the recurrent neural network, long short-term memory, and the GRU models with the conventional algorithm. The predicted travel time of each model also was validated by comparing it to the individual probe vehicle data. The value of travel time analysis was also performed to examine the applicability of the model in urban planning and policy making. It was found that the spatio-temporal GRU predicted link and route travel times were most accurate among the four models.