基于特性矩阵分层分析的短期母线负荷预测坏数据处理策略  

Bad Data Processing Strategies for Short-Term Bus Load Forecasting Based on Stratified Analysis of Characteristic Matrix

作　　者：孙谦[1] 姚建刚[1] 金敏[1] 杨胜杰 匡少林 徐振超 

机构地区：[1]湖南大学电气与信息工程学院,长沙410082 [2]湖南湖大华龙电气与信息技术有限公司,长沙410012

出　　处：《电工技术学报》2013年第7期226-233,共8页Transactions of China Electrotechnical Society

基　　金：国家高技术研究发展计划(863计划)项目(2008AA042802);国家重点产业振兴和技术改造项目([2010]2272)

摘　　要：原始数据分析是提高短期母线负荷预测精度的重要环节,提出一种基于特性矩阵分层分析的坏数据处理策略。首先研究划分样本集最优簇结构的聚类算法。利用AP聚类计算标幺曲线样本集的聚类数搜索区间;从大到小排列各样本点的密度指标,得到初始化矩阵;通过Silhouette指标进行有效性检验,最终得到最优聚类结果。参照特征曲线,计算反映负荷点性质的横向及纵向特征向量,进而形成特性矩阵。运用判别准则对日负荷曲线的特性矩阵进行分层分析,并针对不同变化特性的母线负荷制定相应的坏数据处理策略。实例分析表明,本文提出的方法可有效改善原始数据的质量,提高母线负荷预测精度。As an important link, the original data analysis would improve the accuracy of short-term bus load forecasting a lot. Thus, a bad data processing strategy based on stratified analysis of characteristic matrix is presented. Firstly, the AFS clustering algorithm for dividing sample set optimal clustering structure is studied. The search interval of clustering number for per-unit curve sample set is calculated using AP（Affinity propagation） clustering algorithm, and the initialized matrix is obtained on the basis of the density index arranged according to the decreasing order. Then the optimal clustering results are finally achieved by effectiveness testing based on the Silhouette index. Referring to the characteristic curves, the horizontal and vertical eigenvectors reflecting properties of the load points are calculated, and the characteristic matrix is formed. By applying the discriminant criterion, the stratified analysis for the characteristic matrix of daily load curve is carried out, and thereafter the corresponding bad data processing strategies focusing on bus loads which have different variation of characteristics are established.Case study shows that the proposed method could improve the quality of raw data as well as the bus load forecasting accuracy effectively.

关 键 词：短期母线负荷预测 坏数据处理 特性矩阵 分层分析 聚类算法 

分 类 号：TM715[电气工程—电力系统及自动化]