Research On Hot Region Mining Technology Based On Spectral Clustering

With the rapid development of GPS(Global Positioning System)and other positioning technologies,we can more easily and quickly collect rich mobile trajectory data.Hot region mining of mobile trajectory data plays an important role in urban traffic management,road planning and location-based services.Traditional trajectory data mining methods include K-means,DBSCAN and other algorithms,are difficult to select parameters and easy to affect the clustering effect.And the clustering performance is poor in non convex data sets or data sets with uneven density and large difference in clustering distance.In order to solve these problems,this thesis takes the geolife data set published by the Asian Research Institute as the research object,and conducts cluster analysis on the GPS trajectory data included in the data set.The main research work is as follows:1.This thesis explores the adaptive NJW spectrum clustering algorithm,introduces it into the trajectory data clustering,and proposes the hot region mining algorithm based on the adaptive spectral clustering(Hot region mining algorithm based on adaptive spectral clustering,HRMSC).The adaptive scale parameter ?i can reflect the real distribution of the data set more accurately,and the clustering number k can be automatically determined by calculating the eigen gap,which can effectively avoid the bad results due to the selection of parameters.Compared with the traditional K-means and DBSCAN algorithms in the real trajectory data set,the experiment shows that the hrmsc algorithm has superior clustering effect.2.In the process of clustering and mining trajectory data using hrmsc algorithm,K-means algorithm is adopt to complete the final clustering operation.K-means clustering algorithm is sensitive to the initial value,and it is easy to fall into local optimum,which will affect the clustering effect of hrmsc algorithm on trajectory data.Therefore,k-medoids algorithm,which uses variance to optimize the initial center,is used to complete the final clustering operation,and(Hot region mining algorithm based on K-medoids optimized spectral clustering)KDHRMSC algorithm is proposed.Compared with the hrmsc algorithm on the real trajectory data set,the experimental results show that the KD-HRMSC algorithm has a better clustering effect.3.Residents' travel can be divided into two situations: weekdays and weekends.Users' travel stay points are extracted from the trajectory data in different time periods of a day,and KD-HRMSC algorithm is used for clustering.Using Baidu map API to visualize the hot region,analyze and summarize the travel rules of residents in different situations.