Identification of rock discontinuity sets based on a modified affinity propagation algorithm

Identification of rock discontinuity sets is an important foundation for stability analysis of rock engineering applications. A modified affinity propagation (AP) algorithm is proposed for identifying rock discontinuity sets based on discontinuity orientations. The method considers all data points as potential clustering centers simultaneously, which could avoid the hard selection on initial clustering centers as well as achieve the global optimization. Euclidean distance measure in the original algorithm is not suitable for the clustering of discontinuity orientations and therefore a similarity measure based on the negative sine-squared value of the acute angle between discontinuity unit normal vectors was used. Moreover, the Silhouette validity index was introduced to determine the optimal clustering number. The validity of the new method was tested by using artificial data, and in-situ data compiled from the literature. Finally, the proposed method was applied to discontinuity grouping in an underground water-sealed oil storage cavern in Liaoning Province, China. The results show that the new method can effectively filter noisy data and achieve good clustering results with stronger robustness than other methods.