基于D-Vine Copula构建内蒙古四邻近地区风速相依模型

Modeling Dependency Structure of Wind Speed by D-Vine Copula in Four Neighboring Areas of Inner Mongolia

摘要: 采用D-Vine Copula方法测度内蒙古四近邻地区最大风速的关联性，该方法将多元联合分布通过Pair-Copula分解成边缘密度和二元Copula函数的乘积形式。根据Kendall秩相关系数选择最优的D-Vine Copula结构，使用两阶段法求解模型参数。首先构建边缘密度，然后使用逐树估计和联合估计方法估计二元Copula参数并选择最优分布。通过比较赤池信息量(AIC)发现，相比于逐树估计方法，联合估计参数的结果拟合效果更优。模拟发现四近邻地区间风速间存在不同形式的关联性，D-Vine Copula方法能够灵活的测度这种高维随机变量的关联性差异。

Abstract: The D-Vine Copula method is used to measure the correlation of maximum wind speed in four neighboring areas of Inner Mongolia. Based on this method, the multivariate joint distribution is decomposed into a product of the marginal densities and the bivariate Copula functions in terms of the Pair-Copula technique. The optimal D-Vine Copula structure is selected according to Kendall rank correlation coefficient, and two-stage strategy is used to solve the model parameters. The marginal density is fitted first, and then the Pair-Copula function is simulated by the tree by tree estimation and global joint density estimation methods. By Akaike Information Criterion (AIC), it can be seen that the fitting effect of the global joint density estimation is better than that of tree by tree. It is found that there are different forms of correlations between wind speeds in neighboring areas, and the D-Vine Copula method can flexibly measure the correlation of such high-dimensional random variables.