一种顾及邻近域内实体间距离的空间异常检测新方法

摘要

空间异常检测已成为空间数据挖掘和知识发现的一个重要研究内容.空间异常蕴含着许多意想不到的知识,现有的空间异常检测方法大多依据空间邻近域的非空间属性差异来计算偏离因子,忽略了邻近域内空间实体间距离的影响.本文首先讨论了空间邻近域内实体间距离对空间异常检测的影响,在此基础上,提出了一种顾及邻近域内实体间距离的空间异常度量方法--SOM法,并分析了它的复杂度.由于该方法是利用实体非空间属性的加权内插值与实测值的差值作为度量空间异常程度的参数,从而顾及了邻近域内所有实体相互间距离对非空间属性偏离的影响,并且克服了现有检测方法在不均匀分布空间实体集内寻找空间异常的缺陷.最后,通过一个实际算例验证了所提方法的可行性和正确性.%Detection of spatial outliers has been one of the hot issues in the field of spatial data mining and knowledge discovery. So far, the detection of spatial outliers is determined by spatial outlier factor in most of the existing methods, while geometrical distances among their corresponding spatial neighbor are ignored. In this case, these existing methods are inappropriative to the spatial inhomogeneous distribution. To overcome this limitation, this paper presents a new method for spatial outlier detection, named as spatial outlier measure method (SOM for short). At first, some concepts related to the SOM are defined, such as the attribute gradient, the inverse distance weight and the degree of spatial outlier. The algorithm of the SOM is further presented. One can easily find that the new method considers the distances among the neighborhood and their effects on the attribute values of the target entities, and the degree of spatial outlier is used to check spatial outliers. Finally, a practical example is employed to demonstrate the validity of the method proposed in the paper, where the Cr concentration data of soil in a southern city of China are utilized.