联合知识图谱与改进高斯混合模型的电力用户聚类方法

Power user clustering method based on a combination of knowledge graph and modified Gaussian mixture model

随着电力用户数据复杂程度不断增大,为了改善人工处理海量复杂文本数据的低效方式,以及解决现有聚类方法存在的寻优能力差、紧凑性不足的问题,提出一种联合知识图谱(knowledge graph,KG)与改进高斯混合模型的聚类方法,利用KG将复杂的文本拆解成相关的知识元结构,并利用知识库对所需内容进行提取,规范高斯混合聚类模型所需要的输入数据,通过期望最大化(expectation-maximization,EM)迭代出良好的聚类结果,进而挖掘大量用户数据中的潜在信息。案例分析表明:与K-Means和层次聚类等典型聚类方法相比,所提方法具有更好的聚类结果、分类效果和全局寻优性能,验证了所提方法的可行性和有效性。

With the data of power consumers becoming increasingly complicated,this paper proposes a clustering method combining knowledge graph(KG)and modified Gaussian mixture model for the purpose of ameliorating the inefficient way of manually processing massive complex text data,and remedying the poor optimization ability and the lack of compactness in conventional methods.KG disassembles the complex texts into related meta-structures of knowledge,and extracts the required content through knowledge base,which standardizes the required input data of the Gaussian mixture clustering model.A good clustering result is iterated by means of expectation-maximization(EM)algorithm.Additionally,the massive potential user information can be explored.Case analysis indicates that,compared with typical clustering methods such as K-means and hierarchical clustering,the proposed method has better clustering results,classification effect and global optimization performance,which verifies its feasibility and effectiveness.