面向类别不平衡负荷序列模式识别的两阶段选择集成学习策略  

Two-stage Selective Ensemble Learning Strategy Enabling Pattern Recognition of Class-imbalanced Load Series

作　　者：王圆圆 韩丁 王世谦 白宏坤 王磊[2] 刘洋[2] WANG Yuanyuan;HAN Ding;WANG Shiqian;BAI Hongkun;WANG Lei;LIU Yang(Economic Research Institute,State Grid Henan Electric Power Company,Zhengzhou 450052,Henan,China;College of Electrical Engineering,Sichuan University,Chengdu 610065,Sichuan,China)

机构地区：[1]国网河南省电力公司经济技术研究院,郑州450052 [2]四川大学电气工程学院,成都610065

出　　处：《电力系统及其自动化学报》2023年第1期86-95,共10页Proceedings of the CSU-EPSA

基　　金：国网河南省电力公司科技项目(5217L021000C)。

摘　　要：为解决集成学习负荷模式识别中的类别不平衡及基分类器冗余等问题,提出一种计及类别平衡的两阶段选择集成电力负荷模式识别方法。首先,采用一种基于密度聚类的高斯人工合成少数类过采样技术,根据少数类负荷样本的密度分布特性合成新样本,以强化负荷分类模型对少数类负荷样本的学习。然后,设计出一种包括基分类器聚类剪枝和优化选择集成的两阶段选择集成策略,基于基分类器池的训练结果,遴选最优基分类器子集参与负荷分类任务。最后,通过UCI标准数据集算例验证了所提方法的有效性和优越性。To solve the issues of class imbalance and base classifier redundancy in ensemble learning load pattern recognition,a power load pattern recognition method based on two-stage selective ensemble learning was proposed with the consideration of class balance. First,a density clustering-based Gaussian synthetic minority over-sampling technique(DCB-GSMOTE)which synthesizes new samples according to the density distribution characteristics of minority load samples was presented,so as to strengthen the learning of minority load samples by the load classification model. Meanwhile,a two-stage selective ensemble learning strategy consisting of clustering-based pruning and optimization-based selection integration of base classifiers was designed. Based on the training results of a base classifier pool,the optimal subset of base classifiers was selected for the load classification task. Finally,the effectiveness and superiority of the proposed method were verified by examples based on UCI standard data set.

关 键 词：负荷模式识别 类别不平衡 基分类器冗余 选择集成 

分 类 号：TM734[电气工程—电力系统及自动化]