The Approach Of Network-Constrained Point Analysis Based On Spatial Model And Its Application In Motor Vehicle Crash

With the rapid development of information technology and explosive amount of spatial big data, the opportunity always walk along with challenge in the field of spatial analysis. On the one hand, traditional spatial analysis are commonly conducted in large geographical scale that could not satisfy the need for analysis on fine scale data, and spatial analysis based on small scale has become a thriving research field. On the other hand, as a special case of geographical phenomenon, network-constrained point event has been gained increasingly attention from researchers, besides, previous studies focused on the analysis of spatial process and spatial pattern of network-constrained spatial point event have revealed that there were several shortages by applying the traditional analytic paradigm. Therefore, many studies have attempted to develop novel approaches for this network-constrained point event. In this paper, we aimed at network-constrained point events in geographical space and proposed the theory framework for modeling network-constrained point. Besides, the methodology of point pattern analysis in the network-constrained environment has been established in this research. Furthermore, the availability and practicality of this analytical paradigm has been reviewed through simulation experiments and a case study by using motor vehicle crash in Wuhan as an example.In the following paper, the method frame for network-constrained was proposed, especially the imporved and combined method for deteceting local cluster, besides, this proposed spatial models were used for detecting of risker road segments of motor vehicle crash and a mixed method was used to explore the unspatial attribute(factors) of motor vehicle crash, that is, a series of geographical method were used to solve the problem of motor vehicle crash to provide reference for the traffic planning and traffic manegement. Hence, the research in this paper is not only an important branch in geographical area, but also a considerable foucs in interdisciplinary of traffic safety and geography. The dissertation was composed by the following parts:(1)The dissertation made a detailed analysis of the theory background and application background of network-constrained phenomenon and point events, motor vehicle crash and spatial analysis. About the theory aspect, the critical issues and problem have been discussed in detail for the method and models for network-constrained point events from the development of these related studies, such as lack of a systematic model frame of spatial point pattern analysis for network-constrained points. About the application aspect, the application summarized the present application of network-constrained method according to the event type, such as motor vehicle crash point event, crime point event and other point event in city. Then the technology and methods in GIS were proposed, especially, the spatial model frame for network-constranined point analysis was proposed, tested and applied to detect riskier road segments of motor vehicle crash.(2) Method study:proposed basic idea for modeling network-constrained point in space, established the method frame of point pattern analysis for network-constrained point events,especially,so as to the shortage of network kernel density estimate in cluster detecting, integrated method combining network kernel density estimate and network spatial autocorrelation was proposed. The basic idea of spatial models for network-constrained points is extended the traditional spatial analysis methods in 2-dimension space to 1.5-dimension network-constrained space, the major difference between the two is the measure of distance, changing Euclidean distance to network distance. Firstly, point pattern and point process were introduced and modeled, based on this, network-constrainded point, the network-constrainded point pattern, the network-constrainded point process were proposed and modeled.Then, the spatial analysis model frame for network-constrainded points was built and expounded. Finally, according to the different characteristic of research object, this frame was interpreted to first-order effect method and second-order effect method for detecting the point pattern. The method of network nearest-neighbor distance(NNN), network kernel density estimate(NKDE),event-based LINCS,attribute-based LINCS and combined LINCS was proposed and explained with their basic ideas, principle, calculation and implementation. Besides, the availability and performance of these methods with compared with traditional methods and themselves using 2*2 groups of experiments, each with 10 simulutions. It has been shown that network-constrained approaches can avoid the error, besides, as for random pattern all the approaches are availability in some extent and as for cluster pattern all the approaches work well in detecting the clusters, especially, the combined method can get more clusters which is closer to the real value.(2) Practical application:The dissertation took motor vehicle crash as a case study to make a practical application of this spatial model to detect crash clusters and risker road segments.Firstly, traditional methods of identification of risk road segments were discussed, and spatial model for motor vehicle crash was proposed. Then, motor vehicle crash in Wuhan was chosen as case study and introduced with its study area, data source, data preprocessing. Considering that the location of motor vehicle crash were recorded with geographical address, a spatial method using address matching to locate each motor vehicle crash was proposed. In the end, the first-order effect methods and second-order effect methods were applied to detect the cluster and risker road segments. It was found that motor vehicle crash in Wuhan shown as a tendency of cluster pattern in the road network, and the cluster pattern was more significant when network-constrained spatial models were used for analysis, besides, compared with planar spatial models, the network-constrained spatial models which identified finer cluster of motor vehicle crash in road network scale may be a more significant guide for traffic safety. As for traffic safety, a detection of cluster of motor vehicle crashes can be usefull, and in order to make a further decrease of traffic loss, the factors of motor vehicle crash were explored at level of the macroscopic and data mining about the non-spatial attribute of point event, moreover, a mixed method for mining the multi-dimensional and nonlinear non-spatial attribute of point events was expounded and employed.In conclusion, the main content and the problems were summarized and the further studies in this field were presented at the end of this dissertation.