基于多视角低秩稀疏子空间聚类的气候分区

Climate Zoning Via Multi-View Low-Rank Sparse Subspace Clustering

选取中国661个气象台站的部分气象数据,采用多视角低秩稀疏子空间聚类方法对其进行区划。先计算相对湿度、大气压、日照时数、日平均温度和气温日较差之间的最大信息系数,结果表明所选取的气象要素不存在较强的相关性,从而有利于多视角聚类。再分别基于线性核函数和高斯核函数,提出对所选气象台站进行多视角低秩稀疏子空间聚类的方法。最后将所提方法与k均值聚类进行实验对比,结果表明,上述方法对中国气候分区更有效。

This paper selected part of the meteorological datafrom 661 meteorological stations in China,and employed a multi-view low-rank sparse subspace clustering method to partition them.Firstly,the maximum information coefficients were calculated among relative humidity,atmospheric pressure,sunshine hours,daily average temperature and diurnal temperature range.The results show that the chosen meteorological elements do not have strong correlation,which is conducive to multi-view clustering.Then,based on the linear kernel function and the Gaussian kernel function respectively,a method of multi-view low-rank sparse subspace clustering was proposed to classify the given meteorological stations.Finally,the experimental comparison was made between the proposed method and k-means clustering,and the experimental results validate the former is more effective in performing the climate zoning of China.