An Effective Selection Method for Vehicle Alternative Route under Traffic Congestion

Abstract

The development of the city transportation system provides us lots of conveniences, but it also can brings traffic congestion causing environmental pollution and increasing the travel cost. The current research mainly focuses on traffic volume prediction, route recommendation. However, it is also very mea-ningful and instructive to select a suitable less time-consuming alternative route on a specified congested road. In general, skilled taxi drivers who are relatively familiar with the traffic condition would like to choose the less time routes to avoid peak congestion road segments. Inspired by above ideas, in this paper, we propose a novel hybrid framework which integrates both urban traffic flow characteristics theory and machine learning techniques. We first describe the problem definition, then capture a typical set of congestion road features from the GPS trajectories, the features include traffic volume, road speed limit, route distance, traffic light, and weather features. After that, the most commonly used top-k candidate alternate routes based on historical data are generated, then the feature representations for congestion are feed to train the deep learning model, and the best alternative route is selected after the training process. Extensive experiments on realistic datasets derived from realistic car services demonstrate the superiority of our methodologies.