[1]李盟,温伍正宏,潘甦.基于双缓冲区的概念漂移检测方法[J].计算机技术与发展,2025,(03):103-108.[doi:10.20165/j.cnki.ISSN1673-629X.2024.0354]
　LI Meng,WEN Wu-zheng-hong,PAN Su.Concept Drift Detection Method Based on Double Buffers[J].,2025,(03):103-108.[doi:10.20165/j.cnki.ISSN1673-629X.2024.0354]
点击复制

基于双缓冲区的概念漂移检测方法()

《计算机技术与发展》[ISSN:1006-6977/CN:61-1281/TN]

卷:
期数:
2025年03期
页码:
103-108
栏目:
人工智能
出版日期:
2025-03-10

文章信息/Info

Title:
Concept Drift Detection Method Based on Double Buffers
文章编号:
1673-629X(2025)03-0103-06
作者:
李盟; 温伍正宏; 潘甦
南京邮电大学 物联网学院,江苏 南京 210003
Author(s):
LI Meng; WEN Wu-zheng-hong; PAN Su
School of Internet of Things,Nanjing University of Posts and Telecommunications,Nanjing 210003,China
关键词:
概念漂移; 双缓冲区; 在线序列极限学习机; 漂移检测机制; 不确定数据流
Keywords:
concept drift; double-buffer; online sequential extreme learning machine; drift detection mechanism; uncertain data streams
分类号:
TP301
DOI:
10.20165/j.cnki.ISSN1673-629X.2024.0354
摘要:
在数据分析中概念漂移问题是经常发生的,这导致了模型不能适应数据分布的动态变化。 针对如何处理流数据中的概念漂移这一问题进行了研究,以提高数据分析性能。 为此,在在线序列极限学习机( OS-ELM) 与漂移检测方法(DDM)结合(DDM-OS-ELM)的基础上,提出了双缓冲区(缓冲区 A 和缓冲区 B)方法。 DDM-OS-ELM 通过结合漂移检测机制和在线序列极限学习机来处理概念漂移,这种方法在检测到概念漂移时就会触发模型更新,在检测过程中,通过双缓冲区解决概念漂移的问题。 缓冲区 A 是解决发生概念漂移后数据量不足导致无法重新训练模型这一问题;缓冲区 B收集发生概念漂移后的数据,使模型适应概念漂移后的数据分布。 实验结果表明,利用双缓冲区不仅可以减少模型更新次数,还提高了模型预测的精度。
Abstract:
In data analysis,the issue of concept drift occurs frequently,which causes models to fail to adapt to the dynamic changes in data distribution. Research has been conducted on how to handle concept drift in data streams in order to improve data analysis performance.Therefore,based on the combination of Online Sequential Extreme Learning Machine (OS-ELM) and Drift Detection Method (DDM-OS-ELM),a Double-Buffer approach (Buffer A and Buffer B) is proposed. DDM-OS-ELM deals with concept drift by combining drift detection mechanism and online sequential extreme learning machine. This method triggers model update when concept drift is detected. In the process of detection, the problem of concept drift is solved by double buffers. Buffer A addresses the problem of insufficient data for retraining the model after a concept drift occurs. Buffer B collects data after the concept drift,allowing the model to adapt to the new data distribution. Experimental results show that the use of Double Buffers not only reduces the number of model updates but also improves the accuracy of model predictions.

相似文献/References:

[1]刘俊杰,张昕,杨乐,等.基于 DELM 的不确定数据流分类算法[J].计算机技术与发展,2019,29(03):101.[doi:10.3969/ j. issn.1673-629X.2019.03.022]
　LIU Jun-jie,ZHANG Xin,YANG Le,et al.An Uncertain Data Stream Classification Algorithm Based on Distributed Extreme Learning Machine[J].,2019,29(03):101.[doi:10.3969/ j. issn.1673-629X.2019.03.022]
[2]姚永生,董育宁,邱晓晖.基于相似性度量的网络流分类模型融合[J].计算机技术与发展,2021,31(12):7.[doi:10. 3969 / j. issn. 1673-629X. 2021. 12. 002]
　YAO Yong-sheng,DONG Yu-ning,QIU Xiao-hui.Network Traffic Classification Model Fusion Based on Similarity Measurement[J].,2021,31(03):7.[doi:10. 3969 / j. issn. 1673-629X. 2021. 12. 002]

更新日期/Last Update: 2025-03-10