Research On Network Spatiotemporal Analysis And Multi-constraint Spatial Regionalization For Urban Traffic Collisions

With the increased mobility due to rapid socioeconomic developments,traffic collisions are happening more frequently in recent years.The traffic safety issues caused by international and domestic traffic collisions are extremely severe.The spatiotemporal analysis of traffic collisions is becoming a hot topic for researchers,as well as regionalization for effective management,which is an important issue in spatial optimization.Over the past decade,many research has been conducted on spatiotemporal clustering patterns analysis and spatial regression analysis of traffic collisions.Fruitful results were achieved,while many problems remain.In addition,there are relatively few studies on the spatial optimization issue for effective management of traffic collisions.This paper focuses on the spatio-temporal analysis and regionalization problem of traffic collisions,aiming to solve the problems in present studies.First,the spatio-temporal distribution of traffic collisions along the road are examined and the relationships to various variables are estimated and analyzed.Based on the analysis results,the multi-constrained network regionalization on traffic collisions is studied.The automatic regionalization on traffic collisions could assist the traffic administration for effective management of traffic accidents.The works include:1)Road network constrained spatio-temporal clustering analysis of collisionsThis study used multi-source and heterogeneous spatio-temporal traffic big data to study the spatio-temporal clustering patterns of traffic collisions from multiple perspectives,which reveals the spatial autocorrelation and its temporal aspect.In particular,we proposed the road network constrained traffic collisions kernel density estimation with both spatio-temporal property and semantic attributes,which identified the hot zones of traffic collisions.Second,the spatial aggregation between POIs and traffic collisions was examined by using Network cross K-function,which discovers POI types that are significantly clustered with traffic collisions.Third,we expanded the differential local Moran's I into the network space,and used it to analyze the changing patterns of traffic collisions over time,and explore the change in the number of collisions between any two time periods and its spatial autocorrelation.Lastly,we extended the local indicators of mobility association to network space,which provides straightforward and quantitative measures of the hotspot changes with traffic collisions.In general,comprehensive analysis is conducted on the spatio-temporal clustering patterns of accidents from multiple perspectives,which could provide decision support for traffic management decisions.2)Road network constrained regression analysis on traffic collisions based on multi-source and heterogeneous spatio-temporal traffic big dataUnlike traditional regression analysis on traffic collisions,this work takes advantage of the multi-source and heterogeneous spatio-temporal traffic big data.The relationships between traffic collisions and various factors are estimated and analyzed.In detail,we first select and preprocess independent variables,such as traffic jam,traffic violation and POI to the traffic collisions regression analysis.The independent variables higher with collinearity are removed through factor collinearity detection,which simplifies the model complexity and reduces the overfitting issue.The second is the global regression analysis of traffic collisions based on the OLS method.Through the stepwise regression,the fitness results were optimized.The third is regression analysis based on SEM,SLM,GWR,and network GWR,which helps to find out the influence of spatial autocorrelation and spatial heterogeneity on the result of regression.The road network constrained GWR method reveals the influence of road network constraint on the result of regression.The fifth is to analyze and compare the experimental results of all the above methods,evaluating the fitting accuracy and complexity,analyze the contribution and significance of the impact factors,and explains the spatial nonstationary property of the impact factors.3)Multi-constraints road network regionalizationBased on the spatio-temporal clustering model and the genetic analysis conclusion of traffic collisions,this paper proposed the multi-constrained road network spatial optimization which produces the automatic regionalization in the network space with multiple constraints such as threshold constraint and boundary constraint.First,this paper discusses the formal definition of the network max-p region model with boundary constraints of the management area and discussed the model parameters,objective functions,decision variables,and constraint conditions.The heuristic algorithm based on tabu search is then elaborated which completes local spatial optimization.The results produced from the above methods are then compared and analyzed.The experiments under different parameter conditions were conducted to verify that the regionalization results produced from multi-constraints road network max-p model are more appropriate for the effective management of traffic collisions.4)The prototype system of spatio-temporal analysis and regionalization for road traffic collisions managementBy integrating the multi-perspective road network constrained spatio-temporal clustering analysis of collisions,road network constrained regression analysis of collisions,multi-conditional-constrained road network spatial optimization of collisions,we designed and implemented the extraction,database creating,storage,analysis,and visualization of multi-source heterogeneous spatio-temporal traffic big data.The prototype system includes the multi-perspective spatio-temporal aggregation analysis module,the collisions road network constraints genetic regression analysis module,and the multi-condition road network constrained space partition module were also implemented.