文献类型：期刊文献

英文题名：Plant-wide block-wise monitoring based on an inter-unit attention parallel convolutional autoencoder

作者：Li, Jian[1];Tang, Yue[1];Yan, Xuefeng[1]

机构：[1]East China Univ Sci & Technol, Key Lab Smart Mfg Energy Chem Proc, Minist Educ, Shanghai 200237, Peoples R China

年份：2026

卷号：37

期号：4

外文期刊名：MEASUREMENT SCIENCE AND TECHNOLOGY

收录：;WOS:【SCI-EXPANDED(收录号:WOS:001677497300001)】；

基金：The authors are grateful for the support of Key Projects of the National Natural Science Foundation of China (62433004).

语种：英文

外文关键词：process monitoring; attention mechanism; convolutional autoencoder; mechanism block; global coupling

摘要：In complex process industries, while global modeling can capture overall trends, it often misses critical local information. Conversely, block-wise modeling excels at highlighting local features but struggles to capture inter-block dependencies. To tackle the challenge of balancing global and local information capture, we introduce a block-wise monitoring method utilizing an inter-unit attention parallel convolutional autoencoder (IUA-PCAE). First, the method partitions process variables into blocks based on the technological structure, introducing self-attention and cross-attention mechanisms to extract both intra-block dependencies within each block and inter-block interactions between different blocks simultaneously. To address long-distance interactions, we design a global fusion block that aggregates information from multiple blocks. Second, a parallel convolutional autoencoder is employed to facilitate the unified reconstruction of both local and global features. Finally, Bayesian inference is used to fuse the reconstruction errors from all variable blocks, forming a multi-level monitoring framework of 'local dependency, global coupling, and decision fusion.' Validation results from numerical simulations, the TE process, and a wastewater treatment system show that the proposed IUA-PCAE surpasses existing methods in terms of fault detection rate and false alarm rate. Moreover, the proposed IUA-PCAE effectively identifies key faulty variables, confirming its robustness and interpretability for monitoring complex industrial processes.