基于Stacking模型融合的专变用户电费回收风险识别方法  

Identification method of electricity charge recovery risk of specialized transformer user based on Stacking model fusion

作　　者：潘国兵[1] 龚明波 贺民 邬程欢 唐小淇 杨吕 欧阳静[1] PAN Guobing;GONG Mingbo;HE Min;WU Chenghuan;TANG Xiaoqi;YANG Lü;OUYANG Jing(Institute of Distributed Energy and Microgrid,Zhejiang University of Technology,Hangzhou 310012,China;Ningbo Power Supply Company of State Grid Zhejiang Electric Power Co.,Ltd.,Ningbo 315000,China;Zhejiang Huayun Information Technology Co.,Ltd.,Hangzhou 310012,China)

机构地区：[1]浙江工业大学分布式能源与微网研究所,浙江杭州310012 [2]国网浙江省电力有限公司宁波供电公司,浙江宁波315000 [3]浙江华云信息科技有限公司,浙江杭州310012

出　　处：《电力自动化设备》2021年第1期152-158,共7页Electric Power Automation Equipment

基　　金：国家重点研发计划项目(2017YFA0700300);浙江省重点研发计划项目(2018C01081)。

摘　　要：针对当前电力公司面临的专变用户电费回收风险,提出一种基于Stacking模型融合的专变用户电费回收风险识别方法。对专变用户数据进行特征处理、特征构造与特征筛选,从样本分布和特征属性上优化模型的泛化性能;利用Stacking模型融合多个基学习器,构建专变用户电费回收风险识别模型。实验结果表明,相较于其他常用的分类算法,所提方法具有更优的精确率、召回率、P-R调和均值、AUC值以及模型泛化性能,对专变风险用户的识别率也更高。Aiming at the current electricity charge recovery risk of specialized transformer users faced by electricity companies,an identification method of electricity charge recovery risk of specialized transformer users is proposed based on Stacking model fusion.Feature processing,feature construction and feature selection are carried out for the data of specialized transformer users,and the model generalization performance is optimized from the sample distribution and feature attributes.Stacking model is used to integrate multiple base learners to construct an identification model of electricity charge recovery risk for specialized transformer users.The experimental results show that,compared with other commonly used classification algorithms,the proposed method has better precision,recall rate,P-R harmonic mean value,AUC(Area Under Curve)value,and model generalization performance,and has higher identification rate of specialized transformer risk users.

关 键 词：专变用户 电费回收 风险识别 Stacking模型融合 LGBM 

分 类 号：TM73[电气工程—电力系统及自动化]