Key Techniques Of Intelligent Travel For Urban Transportation On Road Networks

With the development of mobile internet and global positioning technology,more and more services and application can not only facilitate user's travel,but also collect the travel data of massive users.How to use these data to solve travel problems on urban road network has become one of the most popular research topics.However,the complex spatio-temporal and statistical characteristics of travel data on urban road networks cause many challenges,which includes:(1)the large amount of data leads to low efficiency when making similar matching for travel routes in the spatial dimension;(2)the inefficient usage of spatio-temporal information such as travel trajectory leads to low accuracy when estimating travel time in the temporal dimension;(3)the various spatio-temporal correlations make it difficult to accurately predict the trend of traffic in the statistical dimension.Therefore,this thesis aims to solve the above challenges as follows:1.Travel route matching based on sequence similarity search and join: Travel route matching refers to matching similar travel routes,such as search matching and join matching.Firstly,considering that existing similarity measurements ignore the road network constraint,we propose a new similarity function with regarding each travel route as a sequence of road segments.Secondly,we design a ”filtering-refine” framework based on the technique of prefix signatures,which can help quickly prune most dissimilar data in the ”filtering” stage.Finally,we develop a distributed framework based on Spark.Also,we design some distributed search and join algorithms to support route matching.The results of many experiments show that the matching speed of our method outperforms existing methods by 1 to 3 orders of magnitude.2.Travel time estimation based on trajectory modeling: Travel time estimation means estimating the used time from an origin location to a destination.Considering massive historical trajectory data are useful for the estimation travel time,we try to model trajectories when design the estimation model.Specifically,given a trajectory,we first encode and fuse its spatial and temporal features on road segments,and then encode its sequential features.In addition,to address the issue of missing trajectories in the actual estimation situation,we design a model to encode the origin-destination information,where the model is learned by the assist of historical trajectories.The results of many experiments show that our method can improve the accuracy by more than 30% compared with existing methods.3.Traffic Trend Prediction based on spatio-temporal correlation modeling:Different kinds of traffic data(such as traffic flows and travel demands)are mainly obtained by counting users' trajectories,so there exist strong correlations among these traffic data.Hence,we try to exploit these correlations to jointly predict trends of different traffic data,which aims to improve the prediction accuracy.Specifically,we take three kinds of spatio-temporal correlations into account: the correlation of spatial regions,the temporal periodicity of traffic data,and the correlation among different types.We leverage different deep learning techniques to capture different correlations.The results of many experiments show that our method can improve the accuracy by more than 15% compared with existing methods.