Novel Temporal Autoencoder Model Based on STD for Industrial Soft Sensing

Data-driven soft sensing is widely adopted for real-time quality variable detection due to rapid advancements in machine learning. Industrial process data typically exhibit high dimensionality and strong temporal dependencies, making valuable information extraction essential for soft sensor accuracy...

Ausführliche Beschreibung

Gespeichert in:
Bibliographische Detailangaben
Veröffentlicht in:IEEE transactions on industrial informatics S. 1 - 10
Hauptverfasser: He, Yan-Lin, Jiang, Yu, Xu, Yuan, Zhu, Qun-Xiong
Format: Journal Article
Sprache:Englisch
Veröffentlicht: IEEE 2025
Schlagworte:
Autoencoders
Data mining
Feature extraction
Industrial process
Prediction algorithms
process modeling
Real-time systems
seasonal-trend decomposition (STD)
seasonal-trend decomposition-based temporal autoencoder model (STD-TAEm)
Shape
Soft sensors
Time series analysis
Training
ISSN:1551-3203, 1941-0050
Online-Zugang:Volltext
Tags: Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
Beschreibung
Zusammenfassung:Data-driven soft sensing is widely adopted for real-time quality variable detection due to rapid advancements in machine learning. Industrial process data typically exhibit high dimensionality and strong temporal dependencies, making valuable information extraction essential for soft sensor accuracy enhancement. Autoencoders serve as effective unsupervised feature extractors for capturing key process features, but their reconstruction-focused training limits regression prediction performance. To address this limitation, we propose a novel temporal autoencoder (TAE). This model enhances process data along the temporal dimension through multiple augmentation strategies, enabling extraction of regression-prediction-relevant features by controlling similarity/dissimilarity between enhanced datasets. Furthermore, we develop a seasonal-trend decomposition-based temporal autoencoder model (STD-TAEm) that integrates seasonal-trend decomposition with TAE modules. This integrated approach effectively captures trend and seasonal information, improving regression prediction accuracy. Evaluation using a gas turbine combustion dataset confirms STD-TAEm's effectiveness, with comparative experiments demonstrating superior accuracy over existing industrial soft sensing methods.
ISSN:1551-3203
1941-0050
DOI:10.1109/TII.2025.3618169