判别疑似窃电用户决策树分类模型的研究  

Research on Decision Tree Classification Model for Discriminating Suspected Power Theft Users

作　　者：解慧 张建国[2] 赵富华 赵宇鑫 黄义 张殊溶 XIE Hui;ZHANG Jianguo;ZHAO Fuhua;ZHAO Yuxin;HUANG Yi;ZHANG Shurong(Mathematics Department,Jinzhong University,Jinzhong 030619,China;Physics Department,Jinzhong University,Jinzhong 030619,China;Mechanics Department,Jinzhong University,Jinzhong 030619,China)

机构地区：[1]晋中学院数学系,山西晋中030619 [2]晋中学院物理系,山西晋中030619 [3]晋中学院机械系,山西晋中030619

出　　处：《洛阳理工学院学报(自然科学版)》2025年第1期81-87,共7页Journal of Luoyang Institute of Science and Technology:Natural Science Edition

基　　金：山西省基础研究计划青年科学研究计划项目(202103021223353).

摘　　要：选择用电用户类型、电流周期稳定性和电流变化率三大特征选择作为决策树分类模型的判别条件,由用电用户类型作为决策树的第一特征选择判别条件,识别出非专家样本中的用户可能为疑似窃电用户;采用皮尔逊相关系数确定出用户的电流周期性是否稳定,作为决策树的第二特征选择判别条件,识别出电流周期性不稳定的用户可能为疑似窃电用户;采用最小二乘法确定出用户的电流变化率,拟合出用电用户一天不同时刻的电流变化率情况,作为决策树的第三特征选择判别条件,最终判别出疑似窃电用户和正常用户。利用混淆矩阵和ROC曲线对决策树分类模型进行检验,模型具有良好的稳定性。The three characteristics,including power consumption type,current cycle stability and current change rate,are selected as the discriminant conditions of the decision tree classification model,and the power consumption type is selected as the first feature of the decision tree to identify the non-expert users who may be suspected of power theft users.Pearson correlation coefficient,the second feature selection,is used to determine whether the periodic current of the user is stable.The user with periodic current instability can be identified as a suspected electric thief.The least square method is used to determine the current change rate of the user,and the current change rate of the power user at different times of the day is fitted as the third feature selection criterion of the decision tree,and the suspected power theft user and the normal user are finally identified.The confusion matrix and ROC curve are used to test the decision tree classification model,and it is proved that the model has good stability.

关 键 词：决策树分类模型 皮尔逊相关系数 最小二乘法 混淆矩阵 ROC曲线 

分 类 号：O193[理学—数学]