基于双缓冲区的概念漂移检测方法  

Concept Drift Detection Method Based on Double Buffers

作　　者：李盟 温伍正宏 潘甦[1] LI Meng;WEN Wu-zheng-hong;PAN Su(School of Internet of Things,Nanjing University of Posts and Telecommunications,Nanjing 210003,China)

机构地区：[1]南京邮电大学物联网学院,江苏南京210003

出　　处：《计算机技术与发展》2025年第3期103-108,共6页Computer Technology and Development

基　　金：国家自然科学基金(62071244)。

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。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.

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 

分 类 号：TP301[自动化与计算机技术—计算机系统结构]