The Method Of Mining Frequent Patterns Based On Trajectory Clustering

Mining the frequent patterns of spatiotemporal trajectories is an important research method to analyze the travel characteristics of residents.It has important practical significance and scientific research value for revealing human behavioral patterns,urban planning and addressing the imbalance between traffic supply and demand.Clustering is one of the most important methods of data mining,and it is also a common technical method for exploring frequent areas and identifying frequent routes in cities.How to use clustering technology to mine and analyze frequent patterns based on taxi trajectory data is an important research direction in the field of traffic big data.The measurement of trajectory similarity is the basis of spatiotemporal trajectory clustering technology and an important factor affecting the quality of clustering results.However,most of the existing trajectory sequence clustering methods use point matching to technique measure the similarity between trajectories,resulting in inaccurate measurement and low calculation efficiency,which limits the accurate identification of frequent routes.The clustering methods of trajectory points do not consider the similarity of the surrounding event types and density at the same time,so it is difficult to distinguish the clustering phenomenon formed by multiple natural point processes when extracting frequent activity areas of residents.To solve these problems,this thesis uses deep learning to extract the lowdimensional trajectory representations containing path information to measure the similarity of trajectory sequence data.In addition,the density and type of spatiotemporal events are also considered in the process of clustering trajectory points.Furthermore,a frequent pattern mining framework based on trajectory clustering is established to discover frequent areas and identify frequent paths.Finally,taking taxi trajectory data in Hangzhou,Zhejiang Province as an example,this thesis detects hot spots and high-frequency routes in the city in order to acquire in-depth knowledge in the field of transportation research.The main contents of this thesis are summarized as follows.(1)This thesis expounds the theoretical of trajectory frequent pattern mining from the concept,characteristics,representation and similarity measurement of trajectories,and proposes a trajectory data preprocessing method including abnormal data filtering rules.To improve the accuracy of trajectory similarity measurement and the quality of clustering pattern recognition,a frequent pattern mining framework based on trajectory clustering is established for identifying frequent areas and paths.(2)To simultaneously consider the type and density of trajectory points in the similarity measurement of spatiotemporal events,a clustering method based on the mathematical foundation of multivariate spatiotemporal point process is proposed.This method can robustly distinguish spatiotemporal adjacent clusters with similar density and different proportions of multiple event types under the interference of noise points.Moreover,it is useful for revealing the information behind the clustering patterns formed by various event types,and provides a reference for residents to avoid traffic peak.(3)To solve the problem of inaccurate similarity measurement of trajectory sequence data using point matching,this thesis uses the powerful nonlinear representation extraction ability of deep learning to establish a deep trajectory clustering method that can learn low-dimensional and fixed-length trajectory vectors.A joint training optimization framework that can obtain trajectory representation and clustering centers at the same time is designed,which improves the modeling accuracy and application capabilities of trajectory data.In addition,this thesis takes public and simulated data sets as examples to prove that the method can learn representations from uneven and low sampling rate trajectory data,and improve the quality of clustering pattern recognition.(4)This thesis uses taxi trajectory data in Hangzhou,Zhejiang Province as an example to mine frequent patterns.First,the spatiotemporal characteristics of trajectory data are analyzed.Then,the multivariate spatiotemporal point process clustering method is used to identify frequent areas in the main urban area of Hangzhou at different scales.In addition,the deep trajectory clustering method is used to explore frequent paths between regions of interest.In summary,this thesis expects to make method innovation for trajectory frequent pattern mining,and improve the performance of frequent pattern mining by enhancing the ability of trajectory similarity measurement and clustering method,so as to promote the development of spatiotemporal data mining.