Dynamic Detection Of Urban Hotspots Based On Spatio-temporal Data Stream Clustering

Urban hotspots carry important functions such as transportation,commerce and education,it is of theoretical and practical significance to analyze and extract them.Urban hotspots have both temporal and spatial characteristics,their distribution and intensity change with time continuously.Dynamic detection of hotspots helps government departments to monitor urban traffic and crowd conditions in real time,avoid trampling and other extreme events,and also helps citizens to plan their travel plans rationally.Realizing dynamic detecting of hotspot requires not only continuous updating of data sources,but also incremental learning ability of analysis algorithms.Spatio-temporal data stream is a kind of data source continuously generated and uploaded by spatial positioning equipment,which can provide data support for dynamic detection of hotspots.The clustering algorithm of data stream realizes the dynamic adjustment of algorithm model and clustering result by incremental modification of traditional clustering algorithm.Using spatio-temporal data stream clustering to detect hotspots dynamically and in real-time is reasonable,and in the actual use,we also need to consider some key issue like effectiveness,efficiency and engineering practice.In this paper,we introduce data field theory into classical data stream clustering and propose a new algorithm —— DF-Stream.To verify the effectiveness of the algorithm,we carry out the experiment in Wuhan city and choose taxi data stream as data source.In order to improve the performance,we use distributed computing and multi-threading technology to realizes the multi-granularity parallelization of the algorithm.We use middleware technology and WEB framework to carry out the engineering practice of the algorithm,and choose urban traffic hotspots and people travel hotspots as the main application scenarios for WEB spatial visualization.The main research work of this paper is as follows:(I)Data field based spatio-temporal data stream clusteringData field method can discover and measure the correlation between spatial objects,and combining it with clustering algorithm can effectively detect urban hotspots.However,traditional clustering method can't achieve incremental updates so that it can only analyze historical data.In this paper,the data field theory is introduced into the classical data stream clustering algorithm D-Stream.In the new algorithm——DF-Stream,we replaces density value by potential value and add batch processing mechanism to improve the performance.In the experimental,we simulate the real-time data stream by file and analysis the distribution and evolution patterns of hotspots extracted in different periods of holidays and working days.The effectiveness of DF-Stream algorithm is verified by comparing the result with D-Stream.(II)Multi-granularity parallelization of DF-StreamIn this paper,we design a parallel method with both coarse and fine grained by using distributed and multi-threading technology.Coarse-grained parallelization is mainly realized by distributing and dispatching the task based on a master-slave distributed mode while Finegrained is realized by multi-threading parallelized data field method.Experiment shows that the method can make full use of hardware resources in the cluster.(III)Design and Implementation of WEB System for Dynamic Detection of Urban HotspotsWe design and develop a WEB system for Dynamic Detection of Urban Hotspots which based on DF-Stream algorithm and WEB frameworks.We study the fast processing method of trajectory data stream and design the architecture of the system.Some key technologies such as front-end and back-end frameworks,spatial visualization methods and spatial databases are studied according to the demand of hotspot detection.And we choose urban traffic hotspots and people travel hotspots as the main application scenarios for WEB spatial visualization.