Abstract: Grouping structural planes based on their occurrences is an important way in analyzing the structure characteristics of rock mass. Traditional classification methods usu CUI Xue-jie, YAN E-chuan, CHEN Wu ally greatly rely on geological experience of researchers, which is lack of objectivity. The powerful clustering methods, however, also have drawbacks. In this paper, based on variable length string genetic algorithm, an improved K-means clustering method was proposed, making the automatic clustering of discontinuity occurrence of rock mass available. The essence of the proposed method is to select appropriate initial cluster centers for K-means algorithm applying the genetic algorithm. It overcomes the limitation that the K-means algorithm is usually affected by the initial cluster center and easily converges to the local optimal solution. The application of variable length strings in the improved algorithm of classification, however, can not only automatically determine the number of the optimal structural plane groups during the clustering process, but also provide optimal grouping results. In addition, a new mutation algorithm is proposed based on the occurrence data. It is realized in C++ and is applied in analyzing the occurrence of structural planes in an underground water-sealing rock caverns located at Zhejiang province, China. Reasonable classification results are achieved, proving the applicability of the proposed approach in this paper.

Key words: rock mass discontinuity, occurrence data, K-means algorithm, variable length string genetic algorithm