A Novel Soft Sensor Based on Gating Weighted Multihead Self-Attention LSTM Network for Dynamic Nonlinear Processes
IEEE Transactions on Instrumentation and Measurement · 2026 · 75 · 3514314-3514314
DOI: 10.1109/tim.2026.3693444 ↗Kurz gefasst
Fachartikel in der Zeitschrift IEEE Transactions on Instrumentation and Measurement, 2026. Verfasst von Miao Zhang und 2 weiteren Autor:innen. Die Arbeit nennt 52 Literaturangaben. Gefördert von: National Natural Science Foundation of China, Natural Science Foundation of Zhejiang Province.
Abstract (Originalsprache)
Accurate measurement of quality variables is crucial for effective process monitoring and product quality control. Soft sensors are increasingly being applied to quality prediction in complex industrial processes. However, when the local dynamic nonlinearity of the process data is too strong, the existing soft sensor models still face challenges in extracting multiscale local spatiotemporal features. Therefore, a soft sensor model based on gating weighted multihead self-attention (MSA) long short-term memory network (GWMA-LSTM) is proposed. The network uses multiple MSA mechanisms to extract the spatial features among process variables within time intervals of different scales. In order to make rational use of the temporal nature that still exists among these spatial features, they are guided to the LSTM layer for in-depth temporal feature extraction. Meanwhile, a weighted attention (WA) mechanism layer is designed to dynamically process the feature information integrated by multiple MSA layers, in order to enhance the local key feature extraction ability of the model. Furthermore, the gating mechanism is used to screen the information of the integrated spatiotemporal features, enhancing the model’s processing ability for nonlinear processes. Finally, the experimental results on a numerical example and a sulfur recovery unit (SRU) process demonstrate the effectiveness and superiority of the proposed method.
Förderung
- National Natural Science Foundation of China
- Natural Science Foundation of Zhejiang Province
Zitieren
APA
Zhang, M., Xu, G., & Zhou, L. (2026). A Novel Soft Sensor Based on Gating Weighted Multihead Self-Attention LSTM Network for Dynamic Nonlinear Processes. IEEE Transactions on Instrumentation and Measurement, 75, 3514314-3514314. https://doi.org/10.1109/tim.2026.3693444
Vancouver
Zhang M, Xu G, Zhou L. A Novel Soft Sensor Based on Gating Weighted Multihead Self-Attention LSTM Network for Dynamic Nonlinear Processes. IEEE Transactions on Instrumentation and Measurement. 2026;75:3514314-3514314. doi:10.1109/tim.2026.3693444
BibTeX
@article{Zhang2026,
author = {Zhang, Miao and Xu, Gang and Zhou, Le},
title = {A Novel Soft Sensor Based on Gating Weighted Multihead Self-Attention LSTM Network for Dynamic Nonlinear Processes},
journal = {IEEE Transactions on Instrumentation and Measurement},
year = {2026},
volume = {75},
pages = {3514314--3514314},
doi = {10.1109/tim.2026.3693444}
}
EUVIMED