Research Of Urban Traffic Mining And Destination Prediction Based On Clustering

The rapid development of mobile sensors and network communication,coupled with global positioning system technology,have made the collected mass trajectory become a rich data source for building the urban intelligent transportation system.The research of the urban intelligent transportation system includes the propagation patterns mining of traffic congestion,destination prediction,etc.Amongst of them,propagation patterns mining of the traffic congestion contributes much to travel route planning,and destination prediction dedicates itself to increasing the efficiency of such as taxi dispatching systems,city targeted advertising and so on.Therefore,to improve the existing research work about these two topics,we extend the existing clustering model,proposing a novel ST-CPMC(Spatio-Temporal based Congestion Pattern Mining Clustering)model based on spatio-temporal clustering to min the propagation law of urban traffic congestion and a practical STMAC(Separated Trajectory Movement and Adaptive Clustering)framework based on adaptive clustering for destination prediction.The research of existing propagation patterns mining of traffic congestion seldom involves using the spatio-temporal congestion subgraph to represent the congestion propagation law at macro level.Therefore,we proposes a ST-CPMC model based on spatio-temporal clustering.Specifically,it first performs the clustering process in the spatial dimension based on congestion grids,which produces a congestion subgraph snapshot that covers several clusters of congested grids.Second,it further performs the merge and clustering process in the time dimension based on congestion subgraph snapshots.Considering the particularity of the propagation law of traffic congestion,the merge clustering process is decomposed into three parts: constructing the index adjacency list of congestion subgraph snapshot,cyclically merging the congestion subgraph snapshot and its index adjacency list,and performing the merging process of congestion subgraph snapshots at the level of the cluster of congestion grids,which finally produces a set of spatio-temporal congestion propagation patterns to characterize the spatiotemporal propagation law of traffic congestion.Existing destination prediction works based on deriving probabilities of candidate locations to obtain the destination location explicitly ignore the characteristics of heterogeneous distribution of trajectory data in zone-specific environment.Therefore,we proposes a destination prediction framework based on adaptive clustering,called STMAC.Specifically,it first roughly divides the original trajectory set into several categories according to the trajectory movement trend,and then for each trajectory category,it performs an adaptive clustering process to find fine-grained local clusters as candidate destination locations.Finally,to obtain a more accurate and more reasonable destination location,we adopts the weighted centroid of top-k discovered candidate clusters produced by the trajectory classifier.The experimental result of the propagation pattern mining of traffic congestion on a realworld trajectory dataset shows that ST-CPMC can effectively excavate the propagation law of urban traffic congestion at macro level and locate key congestion propagation grids,which can contribute some useful suggestions to the transportation departments,such as adjusting the measures of traffic control on corresponding road sections.The experimental evaluation and analysis of STMAC and its competitors on two real-word trajectory datasets demonstrate that STMAC achieves significant improvement over competitors in terms of prediction error and accuracy,which indicates that STMAC is applicable to destination prediction in the scenario of heterogeneous distribution of trajectory data in zone-specific environment.