电器与能效管理技术 ›› 2022, Vol. 0 ›› Issue (2): 6-11.doi: 10.16628/j.cnki.2095-8188.2022.02.002

• 研究与分析 • 上一篇    下一篇

基于多分类机器学习模型的智能电表故障预测

李宁1, 张伟1, 郭泽林2, 袁铁江2, 韩鑫磊1   

  1. 1.国网新疆电力有限公司营销服务中心(资金集约中心、计量中心),国网新疆电力有限公司电力科学研究院, 新疆 乌鲁木齐 830000
    2.大连理工大学, 辽宁 大连 116000
  • 收稿日期:2021-09-30 出版日期:2022-02-28 发布日期:2022-03-31
  • 作者简介:李 宁(1968—),男,高级工程师,主要从事电能计量研究。|张 伟(1966—),男,高级工程师,主要从事电能计量与输变电设备运维研究。|郭泽林(1996—),男,硕士研究生,研究方向为智能电表健康管理。
  • 基金资助:
    中央高校基本科研业务项目(人才专项,DUT20RC(5)021);国网新疆电力有限公司科技项目(5230DK20000J)

Fault Prediction of Intelligent Electricity Meter Based on Multi-Classification Machine Learning Model

LI Ning1, ZHANG Wei1, GUO Zelin2, YUAN Tiejiang2, HAN Xinlei1   

  1. 1. State Grid Xinjiang Electric Power Co.,Ltd.Marketing Service Center (Capital Intensive Center,Metering Center),State Grid Xinjiang Electric Power Co.,Ltd.Institute of Electric Power Science,Urumqi 830000,China
    2. Dalian University Of Technology,Dalian 116000,China
  • Received:2021-09-30 Online:2022-02-28 Published:2022-03-31

摘要:

针对智能电表故障数据规模大、维度高、存在错误及异常数据的特点,提出了一种融合多分类机器学习模型的智能电表故障预测方法。采用正态分布补全及箱型图方法,对原始数据集进行缺失值填补及异常值替换;通过计算特征属性与故障类型的相关系数,消除不相关特征,形成特征子集;构建混合采样策略,解决故障数据不平衡问题。计算三种典型机器学习算法处理智能电表故障数据的预测准确率,构建混淆矩阵;计及各分类器的预测能力,构建多分类器融合决策函数。最后,分别采用公共数据集与实际用电数据作为样本,验证了所提方法的有效性。

关键词: 智能电表, 机器学习, 数据预处理, 融合算法, 故障预测

Abstract:

In view of the characteristics of large scale,high dimension,error and abnormal data of smart electricity meter fault data,a fault prediction method of smart electricity meter based on multi-classification machine learning model is proposed.The missing values and outliers are replaced in the original data set by using normal distribution completion and box diagram method.By calculating the correlation coefficient between feature attributes and fault types,the uncorrelated features are eliminated and the feature subset is formed.To solve the problem of unbalanced fault data,a mixed sampling strategy is built.The prediction accuracy of three typical machine learning algorithms to process the fault data of smart electricity meters is calculated,and the confusion matrix is constructed.Considering the prediction ability of each classifier,the multi-classifier fusion decision function is constructed.Finally,the effectiveness of the proposed method is verified by using public data sets and actual electricity consumption data as samples.

Key words: smart electricity meter, machine learning, data preprocessing, fusion algorithm, failure prediction

中图分类号: