Deep spatial-temporal travel time prediction model based on trajectory feature

Abstract

Research on travel time prediction shows its importance in the rational planning of travel arrangements and traffic congestion mitigation. The scale of taxi and online ride-hailing users is huge, and accurate travel time prediction is convenient for passengers to reasonably arrange travel planning. Unlike highways and buses, the trajectory of taxis is complex. At the same time, the travel time prediction of the taxi is affected by many factors. Existing models lack effective trajectory feature. They have low accuracy in predicting travel time in urban areas. This will affect the travel arrangement of passengers. In this paper, we propose a trajectory feature learning method based on the image processing method and time series prediction. Traffic congestion is quantified as the congestion value, and a variety of external factors are considered. The experimental results show that this model has advantages over some classical models in predicting travel time.