Research On Geographically Weighted Regression Modeling Approach Based On Flow Data

With the increasing development of globalisation and informatisation,the flow of various elements has become more frequent and regions have become the space of flows.In this context,the concept of space of flows is gradually replacing the space of place.Mobilities have become a hallmark of modern times.Researchers have begun to use the flow perspective to replace the central place perspective to recognize spatial structure.Transport flow is undoubtedly the most important form of flow to connect and increase communication between different geographical spaces.It impacts and changes the original economic activity space and is a key factor in the shaping of socio-economic spatial structure.Improving understanding of the impact of transport flow on economic activities has wider significance for understanding the spatial distribution law of various economic phenomena.The increasing availability of transport flow data,in the form of big data,enables further incorporation of spatial dependencies and non-stationarity into spatial interaction modeling of transport flows.Currently,the latest geographically weighted regression models are multiscale geographically weighted regression(MGWR)and geographically weighted negative binomial regression(GWNBR).However,the MGWR and GWNBR models only support regression analysis of point data.Although there are some point data research and applications in the GWNBR model,there are no programs available.Therefore,the flow-focused MGWR and GWNBR models are the most important research approach of this paper.The main contents of this article are as follows:Firstly,on the basis of summarizing the current research status,the key technologies in the geographical weighted method are elaborated.It mainly includes calculation of spatial distance,spatial weighted function,type of kernel function,bandwidth optimization criterion,golden section algorithm for bandwidth optimization,variable multicollinearity test,t-test threshold adjustment method,global and local Moran index used for the test calculate the spatial autocorrelation and the test of the correlation coefficient between the variables.In particular,the calculation of spatial weights and the test of spatial autocorrelation are the focus of this paper.Secondly,the definition and calibration process of various geographically weighted regression models are described.A standard data is used to simulate and test these models to verify their advantages,disadvantages,and reliability.The simulation results show that the parameter statistics of the MGWR model is better than the GWR model.The GWNBR model has better model interpretability than GWPR because it takes into account the excessive dispersion of data.Based on the above simulation test results,the expressway transport flow data of Jiangsu Province is taken as an example.The flow-focused global Moran index has been successfully used to measure spatial autocorrelation in expressway transport flow data.The results show that the data is clustered in this area.Then,this paper focuses on the socio-economic determinants of expressway transport flow.This paper uses innovative analysis methods MFGWR and FGWNBR to provide empirical evidence for the interdependence between regional transport and the economy.A comparison of MFGWR and FGWR shows that MFGWR can better explain the heterogeneous process of spatial interaction.Finally,by comparing FGWPR with FGWNBR,it is proved that FGWNBR is a better way to capture the spatial heterogeneity between traffic volumn and explanatory variables.The results show that for excessively dispersed data,the Poisson model will reduce the standard error value of the parameter estimate,which will increase the level of significance of the parameter estimate,resulting in more model bias.Therefore,when the data is excessive dispersion,the Poisson model should be used with caution,and the negative binomial regression model is a very suitable choice.