岩体结构面的聚类分析是进行岩体工程稳定性评价的重要基础。而传统的聚类方法存在对初始聚类中心敏感、分组结果较差且效率低下的问题,因此,本研究提出了一种基于改进麻雀算法(SSA)和K-medoids算法相结合的不连续面聚类算法(KS-SSA)。首先,利用SPM混沌映射初始化麻雀种群;然后,利用改进的SSA算法确定初始聚类中心,以此作为K-medoids算法的初始聚类条件对结构面产状数据进行分组,引入轮廓系数(SC)确定最优聚类组数;最后,通过人工数据和现场实测数据验证了该方法的有效性。并将该方法应用于怒江某岩质边坡的结构面分组。结果表明:新算法对结构面的聚类结果良好,具有较强的鲁棒性。研究成果可以为随机裂隙网络的模拟提供数据支撑。

[Objective] The identification and grouping of rock mass structural planes are essential prerequisites for conducting stability analysis. Existing clustering methods are highly sensitive to initial cluster centers, resulting in suboptimal grouping results and relatively low efficiency. To overcome these limitations, this study proposes a rock mass structural plane clustering method based on an improved sparrow search algorithm (KS-SSA). [Methods] First, SPM chaotic mapping was used to initialize the sparrow population, and then the step factor of the SSA was modified to improve the optimization speed. The improved SSA algorithm was used to optimize the initial cluster centers, achieving a global optimal solution. Subsequently, the K-medoids algorithm was applied for the final grouping of structural plane orientation data. Silhouette coefficient (SC) was used to evaluate the clustering performance. Four sets of artificial structural planes were generated using Monte Carlo simulation to validate the proposed algorithm, and comparative analyses were conducted with classical KPSO, FCM, spectral clustering, and traditional sparrow algorithms. [Results] The differences among different algorithms were primarily concentrated at the boundary points. The proposed algorithm could accurately identify all boundary data points, demonstrating good clustering performance with higher recognition accuracy and better clustering results. Using the silhouette coefficient, the structural planes could be accurately divided into four groups, which was consistent with the actual conditions. Furthermore, the sparrow population size and the number of iterations were identified as two key parameters for clustering analysis. Therefore, based on the artificial data, the optimal population size was determined to be 50 by analyzing the variation of fitness curves. Setting the number of iterations to 50 was appropriate while ensuring computational efficiency. The application of the proposed method to the publicly available Shanley dataset further validated its effectiveness. The new algorithm was applied to 201 structural plane data from a rock slope in the Nujiang River. By calculating the silhouette coefficients, three dominant structural plane groups were identified, which were generally consistent with the field investigation results. In addition, the KPSO algorithm was also employed in the field structural plane data analysis, and computational efficiency of the new method was discussed. The KS-SSA algorithm achieved stable clustering results within just 50 iterations, whereas the KPSO algorithm required 110 iterations for convergence. The runtime of the KS-SSA and KPSO algorithms was 45.97 s and 69.95 s, respectively, indicating that the new method had significantly higher computational efficiency. [Conclusion] The KS-SSA algorithm initializes the sparrow population based on SPM chaotic mapping, effectively preventing the clustering results from falling into local optima. Meanwhile, the step factor of the sparrow algorithm is modified, enabling the KS-SSA algorithm to dynamically adjust the step size and improve the optimization speed. Comparative analysis of three case studies demonstrates that compared with traditional algorithms, the proposed method exhibits superior performance in boundary data identification and improved robustness. This study discusses and clarifies the parameter selection and computational efficiency of the new method. The new method demonstrates promising application prospects and can be applied to large-scale data processing and analysis, providing a theoretical basis for the numerical simulation of three-dimensional networks of rock mass structural planes and rock mass stability analysis.