为了给出客观准确的岩体结构面优势产状，用ISODATA(自组织聚类)算法，采用最小球面距离原则进行分类，将重心向量单位化后作为新的聚类中心，并以类内最大分量标准偏差控制分裂，聚类中心的球面距离控制合并，利用计算机进行编程实现。将此方法应用于湛江水封洞库岩体结构面产状的分析中，通过3组收敛精度依次增加的参数，得出了内聚性越来越高、类间距离越来越小的优势结构面产状，及聚类组数和每组样本数的客观准确结果。研究成果为判别结构面空间发育状况提供方便，弥补了传统图形分析方法不能进行精确划分的不足和以往分析方法不变的聚类组数及原理过程复杂的缺点。

In order to get objective and accurate dominant occurrence of rock mass texture plane, a Self-organized Cluster algorithm is used. The dot belongs to its closed clustering center according to their minimal spherical distance. The unit vector of gravity center is used as the new clustering center. The maximum component standard deviation controls the division, and the minimum spherical distance of clustering center controls the combination. The algorithm is realized by the computer programming. The method is applied to analyze the joint occurrence of underground water-seal cavern in Zhanjiang. By means of three groups of parameters with increasing convergence accuracy, the objective and accurate dominant occurrence, clustering group number, and the dot number in each group are obtained. The result shows that the group cohesion increases and the spherical distances of clustering center decreases. This method is facilitated to discriminate the spatial development of texture plane, makes up for the deficiency of the traditional plots method that cannot give the accurate partition and the shortcoming of the previous method that keeps constant clustering group as well as the complex principle and process.