Research On Trajectory Partitioning And Clustering Technology Based On Node Movement Features

Trajectory data mining has become a hot research field,which has received much attention from many researchers.In Mobile Social Networks(MSN),there is much valuable information hidden in the trajectory data formed by the historical records regarding the node movements,and the trajectory data is closely related with the mobile behaviors of nodes.The potential information can be mined from a large amount of trajectory data through trajectory mining technology,and the results can be applied to the many real scenes.The movements of nodes typically exhibit the regularity and randomness simultaneously,where the regularity is determined by the travel purposes of nodes(e.g.,working,living,business affairs and social activities),and the randomness is reflected in the involuntary movements of nodes(e.g.,the intentions of nodes are switched,which results in the arbitrary direction changes of nodes).All these make the trajectories of nodes very complex,and thus the trajectory data mining task become very difficult.Especially,the trajectory partition and trajectory clustering are the key techniques of trajectory data mining teachnology,which processes the trajectory data and analyzes node movement behaviors.In this thesis,the techniques of trajectory partition and trajectory clustering will be focused and studied.Trajectory partition technique plays a fundamental role in the trajectory data mining.Firstly,the original moving trajectory data is usually very large,and thus a lot of storage space is needed.Then,the trajectory shapes are extremely diverse,which is mainly caused by the limitation of the roads and the random movements of nodes.Therefore,it is a primary task for the problem of analyzing the node movement behaviors.The purpose of the trajectory partition is to remove the redundant data in the trajectories while retaining the vitaltrajectory data,so that the simplified trajectories and the original trajectories are as similar as possible,i.e.,a preferable trade-off between the simplification ratio and partition error is expected to be achieved.This thesis first analyzes the movement behaviors of nodes from the features of moving speeds,stop points and moving directions,and then a trajectory partition method based on multiple movement features is proposed.The method is comprised of three stages.The points where the movement speeds are varied significantify are extracted asthe change points,and then the stay points are extracted according to the time and motion ranges.Finally,the extracted change points and stop points are takenas the feature points,and the Douglas-Peucker algorithm is applied to partition trajectories.The simulation results show that a preferable trade-off can be achieved between the simplification ratio and partition error by the proposed method,and the time consumed for trajectory partition is greatly shortened as well.Trajectory cluster teachnique is to analyze the mobile behaviors of nodes,it essentially applies some existing clustering algorithms to process the trajectory data.In this thesis,trajectory clustering technique identifies the similar trajectory segments which are then classified into some clusters.The trajectory segments in each cluster have the similar moving features,while the segments in different clusters are much different.The results can reveal the potential movement behaviors of nodes,such as the moving directions and moving speeds in a time period.At present,most of the trajectory clustering methods focus on spatial properties or semantic properties.However,spatio-temporal properties are often ignored,and thus some vital mobile information regarding the movement behaviors of nodes is neglected.Therefore,this thesis proposes a trajectory clustering method with spatio-temporal properties,which makes the clustering results have more explicit semantics by using the time properties of the trajectory segments and the position properties.Overall,this thesis proposes a trajectory partition method based on multiple movement features and a trajectory cluster method based on joint spatio-temporal properties,respectively.The simulation results show that the proposed methods can effectively reduce the redundant data in the trajectories and can accurately find the similar movement behaviors of nodes.In the future,more simulations will be tested on other real datasets,and the parameter settings will be optimized to further improve the method performance.