基于LOF的K-means聚类方法及其在微震监测中的应用

矿山微震事件集群是分析矿震的重要参考之一,其准确的划分对矿山微震分布特征和微震活动分析具有重要作用。提出了1种基于局部离群因子(Local Outlier Factor,LOF)的K-means聚类算法并构建了综合SSE评价指标和模型,通过LOF算法检测异常微震事件和选取初始聚类中心,利用Krzanowski-Lai指数确定最佳聚类分组数;采用模拟计算比较了不同数据集大小的聚类效果。结果表明:基于LOF的K-means聚类方法评分最高,聚类结果最好;并利用该聚类方法分析用沙坝矿1649个微震事件的分布特征与微震活动性。实例表明,K=7为最佳聚类分组数,聚类簇的划分受断层滑移和矿山生产活动的影响。

K-means clustering method based on LOF and its application in microseismic monitoring

The microseismic event cluster in mines is a primary reference of mine earthquake analysis, and its accurate division plays an important role for the analysis on microseismic distribution characteristics and micro seismicity of mine. A K means clustering algorithm based on local outlier factor (LOF) was proposed, and the comprehensive SSE evaluation indexes and model were constructed. The abnormal microseismic events were detected and the initial clustering centers were selected by using LOF algorithm, and the optimal clustering number was determined by using the Krzanowski Lai index. The clustering effect of this method was compared with those of the K means clustering in literature ［14］ and the traditional K means clustering by using the simulated calculation. The results showed that the score of K means clustering method based on LOF was the highest, with the best clustering results. The distribution characteristics and micro seismicity of 1 649 microseismic events in Yongshaba mine were analyzed by using this clustering method, which showed that K=7 was the optimal clustering number, and the division of cluster was affected by the fault slip and the mine production.