摘要
在真实的工业物联网环境中,传感器信号常受外界噪声干扰,难以获取纯净数据,这影响了基于数据驱动的时间序列预测任务的准确性.为此,基于改进的对比盲去噪自编码器(Contrast Blind Denoising AutoEncoder,CBDAE)和TCN-Transformer网络,本文提出一种新型时间序列预测框架,称为MoCo-CBDAE-TCN-Transformer.该框架通过引入额外的动量编码器、动态队列和信息噪声对比估计正则化,增强了对时间序列数据动态特征的捕捉能力,并有效利用历史负样本信息.在无需噪声先验知识和传感器纯净数据的前提下,通过捕捉和对比时间相关性和噪声特征,实现传感器数据的盲去噪.去噪后的数据通过TCN-Transformer网络进行时间序列预测.TCN-Transformer网络结合残差连接和膨胀卷积的优势以及Transformer的注意力机制,显著提高了预测的准确性和效率.最后,在公开的四缸过程数据集上进行仿真验证,实验结果表明,与传统的去噪方法和时间序列预测模型相比,本文设计的模型能够获得更好的去噪效果和更高的预测精度,其实时处理能力适合部署在实际的工业环境中,为工业物联网中的数据处理和分析提供了一种有效的技术方案.
In real-world Industrial Internet of Things(IIoT)environments,sensor signals are often contaminated by external noise,making it difficult to obtain clean data,which in turn affects the accuracy of data-driven time series prediction tasks.To address this issue,this paper proposes a novel time series prediction framework named MoCo-CBDAE-TCN-Transformer,which combines an improved Contrast Blind Denoising AutoEncoder(CBDAE)and a TCN-Transformer network.The framework enhances its ability to capture dynamic features of time series data and effectively utilizes historical negative sample information by introducing additional momentum encoders,dynamic queues,and Information Noise-Contrastive Estimation(InfoNCE)regularization.Without prior knowledge of noise or clean sensor data,this framework achieves blind denoising of sensor data by capturing and comparing temporal correlations and noise features.The denoised data are then used for time series prediction by a TCN-Transformer network,which combines the advantages of TCN's residual connections and dilated convolutions,as well as Transformer's attention mechanism,significantly improving prediction accuracy and efficiency.Finally,experimental results show that,compared with traditional noise removal methods and time series prediction models,the model designed in this paper achieves better noise removal effects and higher prediction accuracy through simulation verification on a public four-cylinder process dataset.Its real-time processing capability is suitable for deployment in actual industrial environments and provides an effective technical solution for IIoT data processing and analysis.
作者
许涛
南新元
蔡鑫
赵濮
XU Tao;NAN Xinyuan;CAI Xin;ZHAO Pu(School of Electrical Engineering,Xinjiang University,Urumqi 830046,China)
出处
《南京信息工程大学学报》
北大核心
2025年第4期455-466,共12页
Journal of Nanjing University of Information Science & Technology
基金
国家自然科学基金(62303394)
新疆维吾尔自治区自然科学基金(2022D01C694)
新疆维吾尔自治区高校基本科研业务费科研项目(XJEDU2023P025)。
关键词
去噪自编码器
动量编码器
动态队列
信息噪声对比估计
时间卷积网络
TRANSFORMER
denoising autoencoder(DAE)
momentum encoder
dynamic queue
information noise-contrastive estimation(InfoNCE)
temporal convolutional network(TCN)
Transformer
