Research On The Method Of Extracting And Analyzing Urban Hotspots Based On Trajectory Clustering

The urbanization process affords people morden life. However, it brings challenges and problems, such as air pollution, resource consumption, laggard planning and transportation congestion et al. The traffic congestion has been one primal problem for city development. Urban road networks and the complexity of their surroundings enhance difficulty for solving city transportation problems. The emerging of GPS trajectory data affords new way and method for tackling urban traffic problem. As a common GPS trajectory data, taxi trajectory reflects city traffic condition and records citizens'daily commuting. How to mining implicit knowedge from the massive trajectory data has nowadays become a significant research subject.Urban hotspots refer to those regions with developed business, citizens'frequent commuting and large amount of traffic flow, which to some extent reflect individuals' intensive commuting. Urban hotspots and their spatio-temporal patterns can be, found by combining floating car data with some urban basic geograpthic information data and employing theories and methods such as spatio-temporal analysis and data mining, which is significant for urban traffic guidance and management, location based service et, al. This paper conducts research for detecting and analyzing city hotspots from two perspectives, namely urban transportation and residents'daily commuting.The main work of this paper are as follows:(1)This paper studies multi-pattern urban hotspots detection method based on spatio-temporal correlation. Firstly, method for constructing multi-pattern road network model is studied. Road network model is constructed based on road intersections (point pattern), roads (line pattern) and community (polygon pattern), which helps people know and understand city transportation from different scales. Second, the floating car data are adopted to assess road traffic condition with respect to different time. This paper takes urban traffic flow as an example to conduct analysis. Through correlation analysis it is discovered that city traffic flow displays strong correlation between adjacent time spans. Finally, spatial autocorrlation ststistics are employed to detect hotspots where city traffic flow is aggregately distributed. With global correlation analysis, it is found that city traffic flow displays strong correlation under point, line and polygon pattern and it overall presents aggregate distribution. By contrasting global Moran's I ststistics and global Z scores under different spatial patterns, this paper finds Modifiable Areal Unit Problem commonly exist in spatial auto-correlation of urban transportation and the spatial auto-correlation extent of the traffic flow depends on size of spatial unit. Furthermore, local Moran's I ststistics is used to discover urban hotspots under different patterns and analyze their MAUP effects.(2) From perspective of city road network, this paper studies the influence of road network on city traffic flow under multiple patterns and proposes network centrality index considering road geometric features for deficiency of conventional road network centrality index used for analyzing city traffic flow. Since network structure and function are mutually affected, as a transportation condition on road network, traffic flow is largely influenced by road network. Network centrality can well measure structure of road network. In experiments, conventional network centrality indexes are computed under different granularities, including degree centrality, betweenness centrality, PageRank centrality. The correlation of these indexes with city traffic flow is analyzed. Experiments show the conventional network centrality indexes are deficient for analyzing traffic flow. They only consider network topology and neglect its geometric features. Therefore, this paper proposes and studies network centrality indexes considering road geometric features. Experiments show that the proposed indexes are superior to the conventional ones for analyzing urban traffic flow.(3) This paper proposes a trajectory clustering method based on decision graph and data field and applies it in taxi trajectory data to detect hotspots where citizens frequently travel. This method improves conventional clustering algorithms whose parameters are difficult to select and the number of clusters hard to ascertain. Datafield is adopted to describe the mutural effect between trajectory points and select parameter. Dicision graph is employed to select cluster centers. First, effectiveness of the method is verified by using synthetic data and real data and it is analyzed and compared with other algorithms, including k-means, DBSCAN, single-link, Clustering by fast search and find of density peaks. Experiments show this proposed method can well recognize noisy points, search cluster centers, ascertain the number of classes, find clusters of arbitrary shapes and avoid setting parameters by experience. Most importantly, it is more applicable for trajectory clustering. Furthermore, this method is adopted to detect hotspots from taxi trajectory data. Hotspots on different kinds of days including holidays, workdays and weekends are discovered and their dynamic changes are analyzed and compared during different time spans.(4) Combining trajectory clustering method based on dicision graph and data field with grid division, this paper studies a fast method to detect hotspots where citizens frequently travel. It is employed to discover hotspots of daily commuting. Based on these hotspots, travel flow pattern between them is further analyzed from the perspective of spatial interaction network. Taking into account that efficiency of detecting hotspots with trajectory clustering method based on decision graph and data field mainly depends on the amount of trajectory points, this research conducts grid division for the study area and each grid unit is regarded as a particle. Times of passengers'getting-on and off in each grid unit can be taken as the mass of the particle. Grid units are clustered and therefore, hotspots are detected. So efficiency of this method only depends on the number of grids. Experiments show this method can be effectively applied to detect hotspots from trajectory data of large volume. First, this grid clustering method is employed to detect hotspots corresponding to passengers' getting-on and off points respectively during selected time spans. Then, on the basis of these hotspots, spatial interaction network is constructed. Interaction analysis indexes, including strength, outflow, inflow, net flow ratio, are adopted to analyze interaction pattern between hotspots. Experiments show only a few hotspots have strong spatial interaction with other hotspots, including Wuchang Railway Station, Xudong, Hankou Railway Station, the Optics Valley et, al. Spatial interaction between most hotspots are weak.