基于二次聚类的电力负荷异常数据辨识  

Identification of Abnormal Power Load Data Based on Two Times Clustering

作　　者：王珂[1] 王天秀[2] 

机构地区：[1]东南大学电气工程学院,南京210096 [2]淮安电力公司,江苏淮安223001

出　　处：《电气技术》2014年第11期1-1,2,3,17,共4页Electrical Engineering

摘　　要：历史负荷数据是电力系统进行负荷预测的基础,历史数据异常将会影响负荷预测的准确性和有效性,因此需要对负荷数据进行异常数据辨识。本文以某一节点负荷数据为研究对象,提出一种基于二次聚类算法的异常电力负荷数据辨识方法。运用数据挖掘中模糊聚类算法并结合有效指数准则对负荷曲线进行一次聚类;将一次聚类结果结合神经网络实现对负荷曲线的二次聚类,提取出日负荷特征曲线;根据负荷曲线的相似性和平滑性,辨识异常负荷数据。算例分析结果表明,此方法效果良好。Historical load data is the basis of load forecasting of power system, the abnormal historical data affect the accuracy and effectiveness of load forecasting, hence, it is necessary to identify the abnormal load data. This paper takes a node load data for research object, and presents a method for abnormal load data identification based on two times clustering algorithm, using fuzzy clustering algorithm combining with validity index to cluster the load curve for the first time; using the clustering results combined with neural network to cluster the load curve for the second time and extract daily load characteristic curve; according to the similarity and smoothness of load curve to identify the abnormal load data. The results of bad data identification in examples indicate that the proposed method is feasible and effective. The example analysis result shows that this method is effective.

关 键 词：二次聚类 电力负荷 异常数据辨识 

分 类 号：TM715[电气工程—电力系统及自动化]