空间混频预测模型及其应用研究

发布时间：2017-12-31 06:03

  本文关键词：空间混频预测模型及其应用研究　出处：《重庆大学》2016年博士论文　论文类型：学位论文

  更多相关文章： 空间混频预测 空间预测方法 MIDAS模型 支持向量机

【摘要】：空间混频预测:考虑到区域之间的作用关系紧密(空间相关性问题)和高频数据与低频数据共存(混频数据问题)的预测方法,是大数据时代下经济预测领域不断受到重视的新颖问题。目前相关研究包括空间预测和混频预测两个相对独立的方向,而针对空间混频预测模型及其应用研究较少。鉴于软空间权重矩阵能更贴近现实描述空间关系、最经典的MIDAS为处理混频提供了新视角、支持向量机能有效解决多属性小样本下的多种非线性设定问题,本研究尝试集成上述3个方面的优势,聚焦于空间混频预测模型及其应用研究,提出空间混频预测模型,并将新模型应用于具体空间混频预测问题之中,该过程实现了将空间计量经济学中空间预测、混频数据预测、机器学习方法、软集合理论等多领域知识交叉和融合。主要工作可从以下3个方面论述:第一,既定空间范围内经济变量大多带有一定空间相关性,空间混频预测模型中如何设定空间相关性结构是需首要解决的问题,是构建一元和多元空间混频预测模型的基础。总结和分析常用空间权重矩阵方法后,发现尽管理论上不存在能够描述个体(例如区域)空间关系或空间相关结构的最优空间权重矩阵,但可以发现常用的空间权重矩阵在表述空间距离和位置关系时和真实地理环境的相互关系存在不一致性。这种不一致性包含了结论的不确定性、模糊性甚至是相互矛盾的,直接影响空间混频预测模型的建模。软集合是一种处理不确定性的有效工具,具备属性、参数和映射三要素。而单一空间权重矩阵是一个特殊的软集合。基于软集合的定义和运算可以有效将各种常用的权重矩阵进行融合,拓宽了传统的空间权重矩阵依赖于“距离”的测度,同时考虑邻近区域的影响、非邻近区域的影响,边界相互作用、中心之间辐射作用,构建出符合实际的软空间权重矩阵。给出了基于软集合理论的软空间权重矩阵构建方法的主要步骤,并结合空间计量经济学相关理论给出权重矩阵的满足条件和检验方法。最后,为验证模型的有效性,将软空间权重矩阵应用于区域产业集聚的影响因素分析中,根据一定检验标准,实例表明新方法具备一定的可行性。第二,借助软空间权重更贴近实际描述经济变量的空间相关性的优势、一般预测模型发展到空间预测处理空间相关性的经验、以已有混频预测中MIDAS处理混频建模思路为切入点,构建出基于软空间权重的一元空间混频预测模型。该模型承上启下,是空间混频预测模型的基础模型,对MIDAS预测模型和空间预测模型的初步融合有一定理论意义。首先,总结空间混频预测模型的建模思路,以备借鉴空间混频预测模型引入空间权重的经验,并介绍和分析MIDAS预测模型的基本原理和模型设定方式。其次,鉴于被解释变量和一元解释变量之间频率不一致、同时解释变量显著带有空间相关性,引入基于软集合理论的软空间权重矩阵来修正MIDAS预测模型中混频数据的多项式赋予权重方法,即新模型解释变量的系数由混频数据分布滞后赋权、软空间权重和系数共同决定,其综合反映了带有空间相关性的单一解释变量下混频预测模型的设定方式。构建出考虑一个高频解释变量预测一个低频被解释变量的基于软空间权重的一元空间混频预测模型。而后深入分析新模型的主要特点,提出了检验模型有效性的预测误差或精度指标。最后将所构建新的一元空间混频预测模型应用于中国区域GDP预测之中,通过中国30个省市自治区季度实际GDP增长预测效果分析和中国30个省市自治区GDP区域特征权重分析,证实了模型可行性。第三,基于软空间权重的多元空间混频预测模型是预测领域现实存在的问题,有一定的现实意义和理论价值,但国内外集中解决该问题的研究较少,因而构建出基于软空间权重的多元空间混频预测模型正是本研究的重点。首先,考虑到基于软空间权重的一元空间混频预测模型是在最经典的MIDAS预测模型框架下初步改进、且应用范围有限,以其为参考,综合分析了基于软空间权重的多元空间混频预测模型构建过程中亟待模型解决的问题:解释变量维度增加(变量之间混频)的估计参数增加问题;解释变量维度增加(变量之间混频)的样本量偏小问题;空间混频共存下模型设定非线性设定问题。紧接着,针对性介绍了支持向量机基本原理及其主要优势—核函数拓展模型多元非线性设定和小样本计算优势,通过核函数替代混频预测模型的赋权方式,软空间权重矩阵表征不同频率变量空间相关性,利用最小二乘支持向量回归机构建出基于软空间权重的多元空间混频预测模型。给出了模型参数求最优的方式,并深入分析新模型的主要特点,提出了检验模型有效性的预测误差或精度指标。而后将本章构建新预测模型应用于中国区域生态效率的预测之中,通过中国30个省市自治区生态效率预测效果分析和区域特征分析,证实了模型可行性。
[Abstract]:Spatial prediction: taking into account the mixing area between the role of close relationship (spatial correlation problem) and high frequency data and low frequency data coexist (mixing data) prediction method, is a new problem in the field of economic forecasting under the era of big data has attracted more attentions. The related research including spatial prediction and prediction of mixing two relatively independent direction. The spatial mixing forecasting model and its application research. In view of the soft spatial weight matrix can be more close to the realistic description of spatial relation, the most classic MIDAS provides a new perspective for the treatment of mixing, support vector machine can effectively solve various nonlinear multi attribute small sample set, this study attempts to integrate these 3 aspects of the advantages. Focusing on the spatial mixing prediction model and its application, proposes the space mixing prediction model, and the new model is applied to the concrete mixing space prediction problem In the process the spatial prediction of spatial econometrics, mixing data prediction, machine learning method, soft set theory in areas such as crossover and fusion. The main work can be discussed from the following 3 aspects: first, in certain range of economic variables often has certain spatial correlation, spatial mixing prediction model to set the spatial correlation structure is required to solve the problem first, is to construct a univariate and multivariate spatial mixing prediction model. Summarize and analyze the commonly used spatial weight matrix method, found that although the theory does not exist in the individual description (e.g. area) to optimal spatial weight matrix of spatial relations and spatial structure, but can be found in space the weight matrices used in the expression of the relationship between space distance and location relationship and real geographic environment is not consistent. This inconsistency contains Conclusion the uncertainty, fuzziness and even contradictory, directly affect the model prediction of the spatial mixing. Soft set is an effective tool for dealing with uncertainty, with the attribute parameters and the mapping of three elements. The single spatial weight matrix is a special soft set. Soft set definition and operation can be effective the integration of a variety of commonly used weight matrix based on the measure to broaden the traditional spatial weight matrix depends on the "distance", considering the impact of adjacent area, non adjacent area, boundary interaction between the center of radiation, soft spatial weight matrix in line with the actual construction. The main method of building soft weight space matrix of soft set theory is proposed based on combining to meet the conditions and test methods of spatial econometrics theory gives the weights matrix. Finally, to verify the model The validity, the soft spatial weight matrix is applied to analysis the influencing factors of regional industrial agglomeration, according to certain standards, examples show that the new method has certain feasibility. Second, with the help of the soft space weight advantage closer to the spatial correlation of the actual description of economic variables, the general prediction model to forecast the spatial correlation of the spatial processing experience. The MIDAS has forecast modeling methods of mixing mixing as the starting point, to construct a prediction model of element space mixing soft space based on weight. The model is the basic model of space link, mixing model, has a certain theoretical significance for the preliminary integration of MIDAS prediction model and spatial prediction model. Firstly, summarizes the spatial prediction of mixing the idea of modeling model, for reference space mixing model is introduced and the weights of the space experience, the introduction and analysis of MIDAS based prediction model The principle and model setting method. Secondly, in view of explanatory variables and a variable element between different frequency, and explanatory variables with significant spatial correlation, multi type spatial weight matrix introducing soft soft set theory based on the modified MIDAS prediction model of mixing in data weighted method, the new model to explain the variable coefficient by mixing data distributed lag weight, weight and coefficient of soft space is determined, it reflects the spatial correlation with single variable prediction model under mixing setting mode. To construct a high frequency considering the explanatory variables predict a low frequency dependent variable model predicts a mixing element space soft space based on weight. Then deeply analysis of the main features of the new model, put forward the test the effectiveness of the model prediction error or accuracy index. Finally the construction of a new mixed element space The model is applied in Chinese region GDP forecast, through the analysis and prediction and analysis of 30 provinces and autonomous regions Chinese characteristics GDP weight growth in real GDP Chinese 30 provinces quarter, confirmed the feasibility of the model. Third, multiple space soft spatial weight based on mixed frequency prediction model is the prediction of the existing problems, there are the practical significance and theoretical value, but the domestic and foreign research focus to solve the problem of less so constructed based on multivariate spatial mixing prediction model of soft space weight is the focus of this study. First, considering a mixing element space prediction model of soft space weight based prediction model is improved in the preliminary framework the classic MIDAS, and the application range is limited, as the reference, a comprehensive analysis of the spatial mixing soft spatial weight prediction model in the process of constructing the model based on the solution urgently Question: increase the explanatory variable dimension (variable between mixing) adds to the problems of parameter estimation; increase the explanatory variables (variables between the dimensions of mixing) small sample problem; nonlinear model space mixing coexist set set. Then, according to the introduction of the basic principle of support vector machine and its main advantages: kernel function expansion the model of multivariate nonlinear and small sample set is calculated by kernel function instead of mixing advantage, prediction model of weighting methods soft spatial weight matrix representation of different frequency variable spatial correlation, using the least squares support vector regression to build prediction model of multiple space mixing soft space based on weight. The parameters of this model to find the optimal way, and in-depth analysis of the main the characteristics of the new model, put forward the test the effectiveness of the model prediction error or accuracy index. Then this chapter Gou Jianxin prediction model It is applied to the prediction of regional eco efficiency in China, and the feasibility of the model is confirmed through the analysis of the prediction efficiency and the regional characteristics of 30 provinces and autonomous regions in China.

【学位授予单位】：重庆大学
【学位级别】：博士
【学位授予年份】：2016
【分类号】：F224
，

本文编号：1358455