基于电力客户分群特征的停电敏感度预测算法研究  

On Prediction Algorithm of Blackout Sensitivity Based on Characteristics of Power Customer Clustering

作　　者：罗鸿轩 肖勇 金鑫 LUO Hong-xuan;XIAO Yong;JIN Xin(China Southern Power Grid Research Institute Co.,Ltd,Guangzhou 510663,China)

机构地区：[1]南方电网科学研究院有限责任公司,广州510663

出　　处：《西南师范大学学报（自然科学版）》2020年第10期106-112,共7页Journal of Southwest China Normal University(Natural Science Edition)

基　　金：中国南方电网有限公司科技项目(ZBXJXM20180016)。

摘　　要：基于电力客户分群特征,以广东省某市级供电局全体145.2万客户为研究对象,采用决策树方法对停电敏感度预测算法进行了研究.结果表明,在测试集中,非居民及居民客户的验证集累积提升度曲线及敏感客户累积提升度曲线具有比较接近的变化趋势,这表明决策树CHAID算法模型的普适性较好,在模型中过拟合问题不存在.决策树CHAID算法模型在客户总量上有明显的差别,且在实际停电时住宅客户和非住宅客户群体间的敏感度比例也有很多差别.通过分析决策树CHAID算法模型、稀疏逻辑回归模型、 SVM支持向量机模型3种算法的预测准确率,在居民客户、非居民客户以及全体客户预测准确率中,决策树CHAID算法均高于另外两种模型.Based on the characteristics of power customer clustering,1452000 customers of a city level power supply bureau in Guangdong Province have been taken as the research object in this paper,and decision tree method been used to study the prediction algorithm of blackout sensitivity.The results show that,in the test set,the cumulative improvement curve of non resident and resident customers verification set and the cumulative improvement curve of sensitive customers have a relatively close trend of change,which shows that the decision tree CHAID algorithm model has a good universality,and there is no over fitting problem in the model.The CHAID algorithm model of decision tree distinguishes the total number of customers significantly.There is a significant difference in the proportion of sensitive customers in the actual outage between the residential customers and non residential customers.By analyzing the prediction accuracy of three algorithms,decision tree CHAID algorithm model,sparse logistic regression model and SVM support vector machine model,the decision tree CHAID algorithm is higher than the other two models in the prediction accuracy of residential customers,non residential customers and all customers.

关 键 词：预测算法 决策树 停电敏感度 分群特征 电力客户 

分 类 号：TM73[电气工程—电力系统及自动化]