中国烟草学报2026，Vol.32Issue(4)：31-43,13.DOI:10.16472/j.chinatobacco.2026.T0020
基于对比记忆自编码器的烟机运行状态监控方法
Monitoring method for cigarette machine operating status based on contrastive memory autoencoder
摘要
Abstract
[Objective]This study proposes a contrastive memory autoencoder-based method for detecting the production operating status of cigarette-making machines to improve cigarette production quality.[Methods]Using empty-end data from the Hangzhou cigarette factory as an example,the model was trained on raw key indicators of cigarette rods,including weight,circumference,draw resistance,and empty ends.A contrastive learning framework was used to pretrain an encoder network capable of capturing cigarette-quality features.By amplifying the discrepancy between anomalous and normal data,the method detects production anomalies.In addition,an anomaly ratio and gating parameter were introduced to optimize memory-vector updating and improve model performance.[Results]The model achieved an F1 score,recall,and AUC of 0.9801,1.0000,and 0.9938,respectively,outperforming the current comparison methods and confirming its effectiveness and superiority.[Conclusion]The proposed method does not rely on cigarette-machine fault data.It can accurately and reliably determine cigarette-machine operating status using only normal operating data,indicating substantial potential for practical application.关键词
记忆自编码器/对比学习/异常检测Key words
memory autoencoder/bootstrapped contrastive learning/anomaly detection引用本文复制引用
饶永,包军,余清,陈志兴,潘晓华,叶林辉,翁得鱼,杜晓东..基于对比记忆自编码器的烟机运行状态监控方法[J].中国烟草学报,2026,32(4):31-43,13.基金项目
机理与数据双驱动的卷接机组精益维保研究与应用(No.K1202403009) （No.K1202403009）
