保护样本分布特征的混合采样方法  

Hybrid Sampling Method to Protect Sample Distribution Characteristics

作　　者：张浩 仁璐 阎少宏 ZHANG Hao;REN Lu;YAN Shao-hong(College of Science,North China University of Science and Technology,Tangshan Hebei 063210,China;Hebei Key Laboratory of Data Science and Application,Tangshan Hebei 063210,China;Tangshan Key Laboratory of Data Science,Tangshan Hebei 063210,China)

机构地区：[1]华北理工大学理学院,河北唐山063210 [2]河北省数据科学与应用重点实验室,河北唐山063210 [3]唐山市数据科学重点实验室,河北唐山063210

出　　处：《华北理工大学学报（自然科学版）》2024年第3期122-130,共9页Journal of North China University of Science and Technology：Natural Science Edition

基　　金：河北省自然科学基金面上项目(A2023209002):非线性马氏切换微分系统的离散、间歇随机反馈控制研究。

摘　　要：在大数据窃电检测领域,基于机器学习的异常用电检测方法往往面临数据不平衡问题,影响了模型的泛化性能。为此,提出了一种保护样本分布特征的混合采样算法。首先,根据样本的分布特征提出了密度欠采样和邻域过采样算法。然后,为进一步提高数据处理效果、提升模型性能,给出了失衡度指标来将两种算法结合,并提出了保护样本分布特征的混合采样方法。在两份数据集上进行实验,经过该算法处理的数据集,相较于现有的过采样算法,能够有效减少样本数量,提高模型的训练速度;相较于现有的欠采样算法,能够提高模型准确率和AUC值。In the field of big data power theft detection,abnormal power consumption detection methods based on machine learning often face the problem of data imbalance,which affects the generalization performance of the model.To this end,a hybrid sampling algorithm that protects sample distribution characteristics was proposed.First,density under sampling and neighborhood oversampling algorithms were proposed based on the distribution characteristics of the samples.Then,in order to further improve the data processing effect and model performance,an imbalance index was given to combine the two algorithms,and a hybrid sampling method that protects the sample distribution characteristics was proposed.Experiments were conducted on two data sets.Compared with the existing oversampling algorithm,the data set processed by this algorithm can effectively reduce the number of samples and improve the training speed of the model;compared with the existing under sampling algorithm,it can Improve model accuracy and AUC value.

关 键 词：数据平衡处理 数据增强 混合采样算法 异常用电检测 

分 类 号：TP309.7[自动化与计算机技术—计算机系统结构]