Research On Similarity Analysis Method Of Trajectory Data

The rapid development of global positioning technology and wireless network technology have promoted the generation of big data in vehicle trajectory.These trajectory data constitute spatiotemporal trajectory sequence,which is used to record the time and position of moving objects.The valuable information contained in the trajectory data can be found through similarity analysis and processing.However,the complex and diverse structure of the traffic network,the massive and uncertain spatiotemporal trajectory data and the asymmetry of the traffic trajectory bring difficulties and challenges to the similarity analysis and clustering among the trajectories.How to measure similarity accurately and efficiently and find the law between trajectories are difficult problems to be solved.In order to solve the above problems,this thesis started from the following two aspects: the trajectory similarity was studied by combining time and position;clustering analysis of the trajectory was performed according to the trajectory similarity and the requirements of practical application scenarios.Based on this,this thesis proposed the longest common subsequence method based on trajectory segment and time function(ST?LCSS)and improved density clustering algorithm(DBSCAN).The specific work is as follows.(1)Based on the mass and uncertainty of spatiotemporal trajectory data,an optimized LCSS algorithm is proposed: ST?LCSS.The traditional LCSS method uses the number of common trajectory points to measure the similarity between trajectories,which retains the integrity of trajectories.However,this method relies too much on the selection of time threshold and distance threshold,which leads to the problem of misclassification of similarity.To solve this problem,on the one hand,the problem of trajectory dissimilarity caused by unreasonable time threshold setting is solved by setting time function;On the other hand,the problem of trajectory asynchronous calculation(starting from different time and different place)is solved by calculating the distance between line segments composed of two adjacent points.Based on the above two improvements,an optimized LCSS method is proposed.The experiment of ST?LCSS algorithm proves that this method can improve the accuracy of calculation to a certain extent.(2)Improved DBSCAN algorithm.The traditional DBSCAN clustering algorithm needs to set two initial parameters artificially: Eps and Minpoints,which leads to the great influence of parameter setting on the algorithm.In this thesis,improved DBSCAN algorithm based on Maxmin distance and K-means optimization was proposed according to the trajectory data.Specifically,the algorithm firstly uses the maximum and minimum distance algorithm combined with K-means algorithm to determine the K-means number of categories and K initial clustering centers;Secondly,sample data was clustered by K-means clustering algorithm.Statistical analysis of clustering results shows that the initial radius of neighborhood Eps of DBSCAN algorithm and neighborhood density threshold Minpoints.Among then,DBSCAN algorithm uses ST?LCSS in(1)to calculate the cluster distance method using the ST?LCSS algorithm to measure the similarity between the trajectory.Based on this,the trajectory data was clustered by using DBSCAN.Experimental results show that this algorithm has a better clustering effect on spatiotemporal trajectory.