EUVIMEDEUVIMED entdecken
Crossref

GeoDSTG: A Geometric-Aware Dual-Stream Spatiotemporal Graph Network for Soft Sensing in Industrial Processes

Dazi Li, Xiyao Zhu

IEEE Sensors Journal · 2026-08 · 26 (15) · 23505-23513

DOI: 10.1109/jsen.2026.3703735 ↗

Kurz gefasst

Fachartikel in der Zeitschrift IEEE Sensors Journal, 2026-08. Verfasst von Dazi Li und 1 weiteren Autor:innen. Die Arbeit nennt 30 Literaturangaben. Gefördert von: National Natural Science Foundation of China.

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

Data-driven soft sensing methods have been widely adopted in industrial fields. However, traditional sequential models struggle to process non-Euclidean data, leading to a deficiency in capturing the spatial topology among sensors. Although existing frameworks incorporating graph neural networks (GNNs) attempt to integrate spatiotemporal features, they still encounter critical challenges. On the one hand, attention-based temporal modeling is often insensitive to local transient fluctuations. On the other hand, aggregation mechanisms relying on predefined adjacency matrices lack relative positional awareness and fail to capture implicit dynamic associations. To address these problems, we propose the geometric-aware dual-stream spatiotemporal graph network (GeoDSTG). First, a deep augmented temporal module (DATM) captures multiscale temporal characteristics by coupling multihead attention with depthwise convolution. Second, a dual-stream spatial module (DSSM) is constructed to calculate relative spatial orientations via a geometric-aware graph branch (GAGB) and uncover latent dependencies through a self-learning graph branch (SGB). Furthermore, a temporal positional encoding (TPE) and a spatial positional encoding (SPE) strengthen spatiotemporal identity. Experimental results on the sulfur recovery unit (SRU) and large-scale industrial plant (LSIP) datasets demonstrate that GeoDSTG outperforms state-of-the-art models across all evaluation metrics. The source code for this work is available at https://github.com/Daisy0412/GeoDSTG

Förderung

  • National Natural Science Foundation of China

Zitieren

APA

Li, D., & Zhu, X. (2026). GeoDSTG: A Geometric-Aware Dual-Stream Spatiotemporal Graph Network for Soft Sensing in Industrial Processes. IEEE Sensors Journal, 26(15), 23505-23513. https://doi.org/10.1109/jsen.2026.3703735

Vancouver

Li D, Zhu X. GeoDSTG: A Geometric-Aware Dual-Stream Spatiotemporal Graph Network for Soft Sensing in Industrial Processes. IEEE Sensors Journal. 2026;26(15):23505-23513. doi:10.1109/jsen.2026.3703735

BibTeX
@article{Li2026,
  author = {Li, Dazi and Zhu, Xiyao},
  title = {GeoDSTG: A Geometric-Aware Dual-Stream Spatiotemporal Graph Network for Soft Sensing in Industrial Processes},
  journal = {IEEE Sensors Journal},
  year = {2026},
  volume = {26},
  number = {15},
  pages = {23505--23513},
  doi = {10.1109/jsen.2026.3703735}
}
Kostenlos starten →