基于改进Transformer的离散车间异常检测方法研究  

Research on Discrete Workshop Anomaly Detection Method Based on Improved Transformer

作　　者：刘睿锋 李威 石东东 张则青 邱永峰 LIU Rui-feng;LI Wei;SHI Dong-dong;ZHANG Ze-qing;QIU Yong-feng(Tianjin Intelligent Manufacturing Branch of Offshore Oil Engineering Co.,Ltd.,Tianjin 300450;Hunan Tianqiao Jiacheng Intelligent Technology Co.,Ltd.,Zhuzhou 412007)

机构地区：[1]海洋石油工程股份有限公司天津智能制造分公司,天津300450 [2]湖南天桥嘉成智能科技有限公司,湖南株洲412007

出　　处：《制造业自动化》2025年第4期120-126,共7页Manufacturing Automation

基　　金：湖南省高新技术产业科技创新引领计划(2021GK4008);工信部2023年度智能制造系统解决方案揭榜挂帅项目;湖南省教育厅科学研究项目-重点项目(21A0356)。

摘　　要：为提升离散制造车间检测的精度和效率,并克服传统异常检测方法在多品种、变批量数据处理及时间序列相关性捕捉方面的不足,提出了一种基于量子粒子群算法加权的Transformer-GAN(QPSO-TGAN)模型。该模型利用Transformer结构的生成器学习时间序列数据的正常模式,鉴别器则提取数据内在特征,以区分正常与异常模式。同时,引入量子粒子群算法进行参数优化,提高异常检测能力。实验基于真实离散车间数据,并与KNN、RNN、VLSTM、LSFL、DAGAN及Transformer等模型对比,结果表明QPSOTGAN在准确率、召回率和F1分数上均优于其他方法,展现出卓越的异常检测性能,能够有效地适用于现实离散车间场景中。To improve the accuracy and efficiency of detection in discrete manufacturing workshops and solve the problems such as difficulties in both detecting multi variety and variable batch data and capturing time series existed in the traditional anomaly detection methods,this paper proposes a quantum particle swarm optimization algorithm weighted Transformer GAN(QPSO-TGAN)model for anomaly detection in discrete workshops.In this model,the Transformer in the generator simulates the normal mode of time series data,while the Transformer in the discriminator captures the intrinsic characteristics of time series data to learn the difference between normal and abnormal modes.It is combined with quantum particle swarm optimization algorithm to iteratively optimize parameters and improve the ability of discrete workshop anomaly detection.The proposed model is tested using real data from discrete workshops,and the results are compared with KNN,RNN,VLSTM,LSFL,DAGAN,and Transformer models.The model has higher accuracy,recall and F1 score than the comparison model.Exhibiting an excellent performance in anomaly detection,the model can be effectively applied in discrete workshop anomaly detection scenarios.

关 键 词：离散车间 异常检测 TRANSFORMER GAN QPSO 

分 类 号：TP18[自动化与计算机技术—控制理论与控制工程]