基于电力日志特征的DBSCAN聚类

DBSCAN Clustering Based on Power Log Characteristics

针对电力系统海量日志导致难以进行人工分类的问题,文章根据电力日志的结构化特征,通过建立日志的特征向量,再使用DBSCAN来对日志进行聚类。以国网甘肃省电力公司日志作为数据集,使用该方法进行了检验。与专家人工分类的对比结果表明,该方法聚类结果与专家分类结果一致,且聚类的类簇中数目误差率小于0.3%。同时,从轮廓系数和Rand指数2个指标来看,DBSCAN聚类方法也能较好地适应该数据集。聚类的结果能够有效减少电力系统运维人员的检查工作量。

In view of the difficulty of manual classification caused by massive logs in power system, according to the structural characteristics of power logs, this paper uses DBSCAN to cluster logs by establishing the feature vector of logs. The logs of State Grid Gansu Electric Power Company was used as the data set for testing. The comparison with expert classification shows that the clustering results of this method are consistent with the expert classification results, and the number error rate in the clusters is less than 0.3%. At the same time, from the two indicators of contour coefficient and Rand index, DBSCAN clustering method can also adapt to the data set. The results of clustering can effectively reduce the inspection workload of power system operation and maintenance personnel.