Research On Traffic Hotspot Discovery Based On Real-time Trajectory Flow

With the development of science and technology,social progress,and the continuous improvement of residents’living standards,private cars have become the daily means of transportation for urban residents and even ordinary families.Therefore,more and more developed cities and even developing cities are facing the challenge of traffic congestion.Based on the existing technology,the GPS data of the vehicle can be obtained in real time through the intelligent vehicle terminal.The GPS track data contains extremely rich information,including but not limited to real-time traffic conditions of the current city.Therefore,how to dig out urban hotspots and regions from real-time massive GPS data is very meaningful for the improvement and development of the entire city’s transportation system and for the transportation guidance of urban residents.The popularity of Internet of Vehicles technology,car-to-people,car and car,information sharing between cars and roads is possible.More and more cars,especially private cars,are equipped with in-vehicle intelligent terminals,which can upload GPS track data in real time.The GPS track data including but not limited to vehicle terminal ID,time,latitude and longitude,etc.The GPS trajectory data automatically uploaded by vehicles in urban roads has high positioning accuracy and wide coverage,and can clearly reflect the current running conditions of urban roads.Therefore,this paper takes real-time trajectory stream data as the research object,and focuses on trajectory clustering and trajectory stream clustering.Based on the typical trajectory clustering algorithm framework,a new algorithm idea is proposed,and on this basis,a trajectory stream data clustering framework based on sliding window is proposed.The research content and innovation points of this paper are as follows:Firstly,for the trajectory of the trajectory segment distance in the classical trajectory clustering algorithm TRACLUS,the deficiencies of the time attributes are not considered.On the basis of the spatial distance metric,the time distance is increased,and the time distance is different from the existing one.The metrics in the literature use a measure based on the overlap ratio,which makes the distance threshold between[0,1];in addition,the same method is used for the measurement of the spatial distance.The distance ratio is used as the unit of measurement so that the distance between the two is at the same magnitude;and on this basis,the logistic regression equation is used to normalize the spatial and temporal distances,and finally the distance metric between the trajectory segments is obtained.Experiments show that this kind of measurement can effectively improve the final trajectory clustering accuracy.Secondly,in the trajectory division stage,a method of feature point selection based on decision tree model is proposed for the current trajectory partitioning algorithm based on trajectory distance metric and artificially set steering angle and speed threshold.The historical data training decision tree model determines the trajectory feature points through the decision tree model.It has been proved by experiments that the feature points can be selected by the classification model,which can greatly improve the efficiency and accuracy of the trajectory division.Thirdly,for clustering result clusters,a clustering cluster feature representation based on minimum enclosing rectangle is proposed.Namely clustering results for a cluster,you can use the=(,,,,,₁₂₃,₁₄₅)to represent the characteristics of the clusters,in which the meaning of each attribute is respectively:clustering cluster all orbit segments in the center of linear and clustering cluster all trajectory in the period of the linear and Angle,clustering cluster corresponding to the lower left corner of the MBR,clustering cluster corresponds to the top right-hand corner of the MBR,clustering cluster number and contained in the trajectory of clustering cluster of trajectory in the earliest time and to the latest time.After the experiment,the analysis result data can be obtained:the trajectory trend of the final clustering result can be accurately expressed by the summary structure,and the summary structure also provides a good micro-cluster summary data structure for the subsequent trajectory-based clustering algorithm.Fourthly,a trajectory flow clustering algorithm based on sliding window is proposed.The moving window model is used to retain the recently reached trajectory stream set,and the trajectory stream set in the window is clustered in real time.The online real-time clustering processing is based on the（Trajectory Cluster Summary）structure,and on this basis,the（Time Cluster Feature）structure is proposed to maintain the evolution process of the entire data stream,and a hierarchical relationship is proposed.To maintain and eliminate obsolete and non-critical data in the trace stream.In summary,this paper focuses on trajectory clustering and trajectory stream clustering.By discovering the shortcomings of current trajectory clustering algorithms,the corresponding solutions are proposed.At the same time,the research on trajectory stream data clustering at the current stage is proposed.A new idea is also proposed.Finally,through the real taxi trajectory data,the experiment proves that the idea proposed in this paper has a good performance in clustering accuracy and efficiency.