基于改进随机森林的电力用户欠费风险分析预警  

Arrears risk analysis and early warning of electricity customers based on optimized random forest

作　　者：李晓蕾 魏玲[2] 王忠强 耿俊成 张小斐 Li Xiaolei;Wei Ling;Wang Zhongqiang;Geng Juncheng;Zhang Xiaofei(State Grid Henan Electric Power Company, Zhengzhou 450000,China;Department of Electrical Engineering, Tsinghua University, Beijing 100084,China;Electric Power Research Institute of State Grid Henan Electric Power Company, Zhengzhou 450052,China)

机构地区：[1]国网河南省电力公司,郑州450000 [2]清华大学电机系,北京100084 [3]国网河南省电力公司电力科学研究院,郑州450052

出　　处：《电测与仪表》2019年第9期56-62,共7页Electrical Measurement & Instrumentation

基　　金：首都蓝天培育行动(Z171100000617001);国网河南省电力公司科技资助项目(52170217001H)

摘　　要：针对当前电网企业电费回收风险,提出了一种基于改进随机森林的电力用户欠费风险分析预警方法。首先,针对欠费用户、正常缴费用户的类别分布不均衡问题,采用SMOTE算法优化原始用户样本分布;其次,选择信息值计算各属性与目标类别属性的相关性,进而优化节点属性的选择;然后,针对影响随机森林分类准确率和性能的主要参数:树的规模nTree、叶子节点的最小样本数minLeaf和属性子集的数量K,采用加温模拟退火算法搜寻最优参数组合;最后,采用改进的随机森林算法对用户未来是否欠费进行分析预测,得到潜在欠费高风险用户。将该方法与逻辑回归、决策树等常用分类算法进行了对比分析,结果验证了该方法的有效性。Aiming at the charging risk in power grid companies, a risk analysis and early warning based on optimized random forest algorithm is proposed in this paper. Firstly, the SMOTE algorithm is used to optimize the original user sample set distribution to solve the inhomogeneous distribution of the arrearage users and normal users. Secondly, the information value is selected to calculate the correlation between attribute features and target category, and then, the random selection of node attributes is optimized. Thirdly, for the main parameters that affect the accuracy and performance of the random forest algorithm: tree size nTree , the minimum sample leaf nodes minLeaf and attribute subset size K , the simulated annealing algorithm is adopted to obtain the best combination. Finally, the optimized random forest algorithm is used to predict the future arrears of users, and obtain the potential high risk users. The method is compared with other classification algorithms such as logistic regression and decision tree, and the experimental results show that the proposed method is effective.

关 键 词：电力用户 欠费风险预测 随机森林算法 SMOTE 信息值 参数组合 加温模拟退火算法 

分 类 号：TM73[电气工程—电力系统及自动化]