Research On Trajectory Clustering Algorithm Based On GPS Data

In recent years,with the widespread application of mobile devices with Global Positioning System(GPS),wireless communication functions and various sensors in Intelligent Transportation System(ITS),the location-based services(Location-based services,LBS)applications have become very common.Moving objects generate and collect a large amount of trajectory data over time.The most common one is GPS data of vehicles[1],which store a large amount of valuable information,such as whether taxis have detour behavior,urban population travel rules,the dynamic traffic information,etc.which is of great value and significance to solve the problems existing in urban life.Therefore,more and more experts and scholars have studied and mined these trajectory data.As an important data analysis method in the field of data mining,trajectory clustering technology has become one of the research hotspots of GPS data mining.Trajectory clustering technology is a simple and direct method of acquiring knowledge from trajectory data.The purpose is to assign the trajectories to different groups or clusters according to the similarity,so that the trajectories within each cluster are highly similar,and vice versa,different clusters share the least similarity.Trajectory clustering can be used to discover the laws of motion and behavior patterns of vehicles,obtain hotspot paths,and predict vehicle behavior.There are many existing trajectory clustering algorithms,but there are still many areas to be improved in terms of efficiency and clustering effect.This paper focuses on the research topic of GPS data trajectory clustering algorithm,and carries out the following two studies for existing problems:(1)Aiming at the efficiency problem of trajectory density clustering algorithm research,this paper proposes a fast density clustering algorithm based on trajectory compression(F＿DDBSCAN).First,analyze the specific causes of the efficiency problems of the existing density clustering algorithm;Secondly,in view of the inefficiency of trajectory clustering caused by oversampling,the trajectory compression technology is used to preprocess the trajectory data,which greatly reduces the number of GPS sampling points and improves the operating efficiency of the algorithm;Finally,kd＿Tree is introduced in the Density Detection Dbscan Algorithm(DDBSCAN)to improve the efficiency of partition density clustering.Taking the taxi trajectory data in Nanjing City as an example,the experimental results show that compared with the existing density clustering algorithm and the traditional clustering algorithm,the F＿DDBSCAN algorithm proposed in this paper improves the accuracy and operating efficiency of clustering to a certain extent,which is significantly better than the comparison algorithm.(2)Aiming at the problem that the clustering accuracy and robustness of the trajectory clustering algorithm needs to be improved,a Trajectory Clustering Algorithm Based on Deep Embedding(DETC)is proposed.Compared with the traditional algorithm,the F＿DDBSCAN algorithm proposed above has certain advantages,but the accuracy and robustness still need to be improved.Besides,most of the existing trajectory clustering algorithms focus on the temporal and spatial characteristics of the trajectory,while ignoring the inherent properties of the trajectory itself.The proposed DETC algorithm can well capture the high-dimensional feature information of the trajectory.It takes the trajectory vector,trajectory angle,trajectory intersection,and time information as the feature space.Using Deep Neural Network(DNN)to learn feature representation and cluster assignment at the same time,to improve the accuracy of the entire clustering algorithm to a certain extent.Comparative experiment results show that due to the complexity of the deep neural network,the computational cost of the algorithm is increased,resulting in the operating efficiency of the DETC algorithm being lower than that of other algorithms.However,DETC algorithm is obviously better than several other comparative clustering algorithms in the accuracy and robustness of clustering.