Stacked semi-supervised autoencoder-regularized RVFLNs for reliable prediction of molten iron quality in blast furnace

This paper proposes a novel stacked semi-supervised autoencoder-regularized random vector functional-link networks (RVFLNs) for reliable prediction of molten iron quality (MIQ) in blast furnace (BF) ironmaking. First, in order to accurately describe the importance of different process variables on t...

Ausführliche Beschreibung

Gespeichert in:
Bibliographische Detailangaben
Veröffentlicht in:Neural computing & applications Jg. 37; H. 17; S. 10425 - 10438
Hauptverfasser: Zhou, Ping, Zhao, Peng, Ou, Zihui, Chai, Tianyou
Format: Journal Article
Sprache:Englisch
Veröffentlicht: London Springer London 01.06.2025
Springer Nature B.V
Schlagworte:
Algorithms
Artificial Intelligence
Coal
Computational Biology/Bioinformatics
Computational Science and Engineering
Computer Science
Consumption
Data Mining and Knowledge Discovery
Distributed control systems
High temperature
Image Processing and Computer Vision
Iron
Iron compounds
Ironmaking
Neural networks
Prediction models
Probability and Statistics in Computer Science
Process variables
Regularization
S.I.: Timely Advances of Deep Learning with applications and Data-Driven Modeling
Special Issue on Timely Advances of Deep Learning with applications and Data-Driven Modeling
Sulfur content
Variables
Attention mechanism
Random vector functional-link networks (RVFLNs)
Blast furnace (BF) ironmaking
Stacked semi-supervised autoencoder
Reliable prediction of molten iron quality (MIQ)
ISSN:0941-0643, 1433-3058
Online-Zugang:Volltext
Tags: Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
Beschreibung
Zusammenfassung:This paper proposes a novel stacked semi-supervised autoencoder-regularized random vector functional-link networks (RVFLNs) for reliable prediction of molten iron quality (MIQ) in blast furnace (BF) ironmaking. First, in order to accurately describe the importance of different process variables on the multivariate MIQ indices, an attention mechanism based on feedforward compensation is introduced. The mechanism is embedded at the front end of the stacked semi-supervised autoencoder deep network. Secondly, to extract the deep feature information which is closely related to the prediction target from the process data, a deep network structure composed of multiple semi-supervised autoencoder models is introduced. The output of the last hidden layer is then used as the input of the subsequent prediction model. At the same time, two regularization terms L 1 and L 2 are incorporated into RVFLNs to sparse network output weights and improve the robustness of modeling, aiming to solve the multicollinearity and overfitting problems of the basic RVFLNs. Experiments using the standard dataset and actual industrial data of BF demonstrate that the proposed method has good performance.
Bibliographie:ObjectType-Article-1
SourceType-Scholarly Journals-1
ObjectType-Feature-2
content type line 14
ISSN:0941-0643
1433-3058
DOI:10.1007/s00521-024-10539-x