文献类型：期刊文献

英文题名：Dual decomposition-enhanced integrated deep networks with bidirectional CNN and semi-supervised GRU for multivariate nonlinear time series forecasting

作者：Zhao, Changcheng[2];Ruan, Yuhui[1];Zhai, Jiazi[3];Peng, Haifei[2];Long, Jian[4,5];Hu, Guihua[2]

机构：[1]Soochow Univ, Sch Polit & Publ Adm, Suzhou 215124, Jiangsu, Peoples R China;[2]East China Univ Sci & Technol, Key Lab Smart Mfg Energy Chem Proc, Minist Educ, Shanghai 200237, Peoples R China;[3]China Petr Pipeline Bur Co Ltd, Reserve Oil Management Serv Co, Langfang, Peoples R China;[4]East China Univ Sci & Technol, Int Joint Res Ctr Green Energy Chem Engn, Shanghai 200237, Peoples R China;[5]East China Univ Sci & Technol, Sch Chem Engn, Shanghai 200237, Peoples R China

年份：2026

卷号：735

外文期刊名：INFORMATION SCIENCES

收录：;EI(收录号：20260219902095);WOS:【SCI-EXPANDED(收录号:WOS:001660313600001)】；

基金：This work was supported by National Natural Science Foundation of China (62373155) .

语种：英文

外文关键词：Multivariate time series; Variational mode decomposition; CEEMDAN; Bidirectional convolutional neural network; Semi-supervised gated recurrent unit

摘要：Multivariate time series in industrial processes often suffer from data complexity, nonlinearity, and missing labels, limiting the performance of soft sensors. To address these issues, this paper proposes an integrated deep learning approach to mitigate these challenges. Initially, we apply Variational Mode Decomposition (VMD) in combination with Complete Ensemble Empirical Mode Decomposition with Adaptive Noise (CEEMDAN) to reduce nonlinearity and complexity in the raw data. Subsequently, we enhance traditional convolutional neural networks by introducing a Bidirectional Convolutional Neural Network (BICNN) for feature extraction, enabling a more comprehensive representation of the decomposed data. Finally, a semi-supervised learning strategy is integrated into the Gated Recurrent Unit (GRU) network, forming the Semi-supervised GRU (SSGRU) model to handle unlabeled data in industrial processes. Additionally, an attention mechanism is employed to further improve prediction accuracy. Experimental results demonstrate that the VC-BICNN-SSGRU algorithm outperforms existing state-of-the-art methods across multiple datasets, showing superior performance in key predictive metrics such as R2, MSE, MAE, RMSE, and MAPE. These findings underscore the algorithm's effectiveness in improving both prediction accuracy and long-term forecasting capabilities for industrial processes, suggesting its potential for broader application in similar domains.