时空数据模型在人口流动研究中的应用

发布时间：2018-09-03 06:02

【摘要】：在人口的迁徙和流动中产生了大量的历史数据，，如何准确高效的利用这些数据得出具有政策导向意义的研究结果尤为重要。有理论表明人口的变动在相邻或者相近的地域之间有较为明显的相关性，但在以往的研究中多是单纯从时间维度上考虑人口结构的变动预测未来的走势或是仅仅对人口的研究在空间统计的范畴内进行分析。本文综合考虑空间和时间的依赖性，针对两类时空数据：连续数据与格数据，分别借助空间残差模型和克里格地理统计模型两个研究方法进行研究。在理论方面，本文主要做了如下工作:对时空数据的类型介绍，时空数据建模的假设与前提，时空模型的形式，参数估计基本思想和极大似然的迭代方法，采用的似然比检验以及预测方法。同时也简要讨论了时空数据模型在实际中的应用；在应用层面借用以上的方法，通过R软件编程实现了整个计算过程。在实证部分中，先对数据进行描述性分析，初步了解其空间和时间维度的分布特点。同时做了统计解释并根据数据的一阶差分特征判定它是适用于空间残差模型的，接下来进行模型拟合和检验发现其空间和时间依赖系数都显著，为人口分布的空间依赖理论提供了数理方面的佐证，并对其进行实际意义的解释。而后对两种模型的拟合结果做出了预测精度和计算效率的对比和评价，发现针对此问题，时空克里格方法在预测精度上优于空间误差模型。以上研究内容为时空数据模型的分析提供数理方面的分析思路。本文在进行实证分析时采用的数据是瑞典的人口数据。因为其空间分辨率高，以教区为地理单位，比省市地区的数据精度要高，有助于在时空分析中得出准确的结论，并且在时间维度上具有完整性和分割一致性。而国内的人口数据主要通过人口普查得到的省市县的数据，在数据的可获得性和空间分辨率上有局限，并且在时空数据模型使用之前，需要对其进行空间化的处理，考虑到工作量和时间成本因素，因而本文直接选取了直接可用来分析的瑞典人口时空数据。但只要数据质量够高，或者数据空间化的预处理已经完成，在遇到时空数据模型问题时，本文的研究思路是值得借鉴和参考的。
[Abstract]:A large number of historical data have been produced in the migration and flow of population. How to use these data accurately and efficiently to obtain policy-oriented research results is particularly important. There are theories that there is a clear correlation between population changes in adjacent or close regions, However, in the previous studies, it is only from the time dimension to consider the change of population structure to predict the future trend, or only to analyze the population research in the field of spatial statistics. In this paper, considering the dependence of space and time, two kinds of spatiotemporal data, continuous data and lattice data, are studied with the help of spatial residuals model and Kriging geographic statistical model, respectively. In theory, this paper mainly introduces the types of spatiotemporal data, the assumptions and premises of spatio-temporal data modeling, the form of spatio-temporal model, the basic idea of parameter estimation and the maximum likelihood iterative method. The likelihood ratio test and prediction method are used. At the same time, the application of spatio-temporal data model in practice is briefly discussed, and the whole calculation process is realized by using the above methods in the application level. In the empirical part, firstly, the data are analyzed descriptive, and the spatial and temporal distribution characteristics of the spatial and temporal dimensions are preliminarily understood. At the same time, the statistical explanation is made and the first order difference characteristic of the data is determined to be suitable for the spatial residual model. Then, the model fitting and testing are carried out and found that the spatial and time dependent coefficients are significant. It provides mathematical evidence for spatial dependence theory of population distribution and explains its practical significance. Then the prediction accuracy and computational efficiency of the two models are compared and evaluated. It is found that the spatial-temporal Kriging method is superior to the spatial error model in prediction accuracy. The above research content provides the mathematical analysis thought for the time-space data model analysis. The data used in this paper are Swedish population data. Because of its high spatial resolution, the parish is a geographical unit, which is more accurate than the data of provinces and cities, which is helpful to draw an accurate conclusion in time and space analysis, and has integrity and segmentation consistency in time dimension. However, the data of provinces, cities and counties, which are mainly obtained from the population census, are limited in terms of data availability and spatial resolution, and they need to be spatially processed before the use of spatio-temporal data models. Considering the factors of workload and time cost, this paper directly selects the space-time data of Swedish population which can be directly used to analyze. However, as long as the data quality is high enough, or the preprocessing of data spatialization has been completed, the research idea of this paper is worthy of reference and reference when we encounter the problem of spatiotemporal data model.
【学位授予单位】：首都经济贸易大学
【学位级别】：硕士
【学位授予年份】：2012
【分类号】：C921

【参考文献】