联合RF-BP-LR的电力客户电费拖欠混合风险预警算法  

Combined random forest-back propagation neural network-logistic regression hybrid risk alert algorithm of power customer arrears

作　　者：谢禄江 蒋荣 皮羽茜 何轶 廖勇[2] XIE Lujiang;JIANG Rong;PI Yuqian;HE Yi;LIAO Yong(State Grid Chongqing Electric Power Company Information&Telecommunication Branch,Chongqing 401120,China;School of Microelectronics and Communication Engineering,Chongqing University,Chongqing 400044,China)

机构地区：[1]国网重庆市电力公司信息通信分公司,重庆401120 [2]重庆大学微电子与通信工程学院,重庆400044

出　　处：《重庆理工大学学报（自然科学）》2022年第5期250-258,共9页Journal of Chongqing University of Technology：Natural Science

基　　金：国网重庆市电力公司科技项目(2021渝电科技8#)。

摘　　要：针对电力客户存在的欠费风险问题,提出一种联合随机森林-反向传播神经网络-逻辑回归(random forest-back propagation neural network-logistic regression,RBL)的电力客户欠费混合风险预警算法。首先,利用随机森林算法对影响电力客户拖欠电费的因素进行一次特征提取;然后,利用反向传播神经网络进行初次风险预测,得到用户对于电费缴纳的信用分值;最后,采用逻辑回归模型进行第二次预测,对用户电费欠缴或拖缴的风险进行有效预警。以某地区的用电客户数据为对象,对比了所提算法与其他预警算法的预测精度,结果表明:所提算法预测精度达到了92.83%,能为电力企业进行用电客户电费管理提供技术支持。Aiming at the arrears risk problem of power customers,a hybrid risk alert algorithm for power customers’arrears is proposed by combining random forest-back propagation neural network-logistic regression(RBL).Firstly,the random forest algorithm is used to extract the characteristics of the factors that affect the power customers’arrears,and then uses the back propagation neural network to make the initial risk prediction,so as to obtain the credit scores of the electricity users for the electricity bills.And finally,the logistic regression model is established to makes the second prediction,which can effectively warn users of the risk of arrears or delayed payment of electricity bills.Based on the data of electricity customers in a certain area,the prediction accuracy of the proposed algorithm is compared with other early-warning algorithms.The results show that the prediction accuracy of the proposed algorithm reaches 92.83%,and the satisfactory effect is achieved,which can provide technical support for electric power enterprises to manage electricity customers’electricity charges.

关 键 词：随机森林 反向传播神经网络 逻辑回归 欠费风险预警 

分 类 号：TP311.5[自动化与计算机技术—计算机软件与理论]