Related Experiment Video
Updated: May 27, 2025
Author Spotlight: Addressing Technical and Subjective Challenges in Measuring Classroom Attention
Published on: December 15, 2023
Soft sensor modeling using deep learning with maximum relevance and minimum redundancy for quality prediction of
Huaiping Jin1, Xin Dong2, Bin Qian1
1Faculty of Information Engineering and Automation, Kunming University of Science and Technology, Kunming 650500, China; The Higher Educational Key Laboratory for Industrial Intelligence and Systems of Yunnan Province, Kunming University of Science and Technology, Kunming 650500, China.
Maximal Relevance and Minimal Redundancy-based Representation Learning (MRMRRL) improves soft sensor quality prediction by enhancing feature extraction and reducing redundancy. This deep learning approach significantly outperforms traditional methods in industrial applications.
Area of Science:
- Machine Learning
- Industrial Process Control
- Data Science
Background:
- Autoencoders (AE) and Stacked Autoencoders (SAE) are popular for soft sensor applications.
- Existing methods suffer from poor feature-quality correlation, information loss, and feature redundancy.
Purpose of the Study:
- To propose a novel Maximal Relevance and Minimal Redundancy-based Representation Learning (MRMRRL) approach.
- To enhance quality prediction accuracy in industrial processes using deep learning.
Main Methods:
- MRMRRL integrates quality-relevant feature extraction, Kernel Principal Component Analysis (KPCA) for redundancy reduction, and an extension layer for information compensation.
- Combines merits from three channels: quality relevance, redundancy reduction via KPCA, and information compensation.
Main Results:
- MRMRRL demonstrated significant performance improvements of approximately 37% and 38% over the baseline SAE in two application examples.
- Outperformed several state-of-the-art deep learning soft sensors.
Conclusions:
- MRMRRL effectively extracts quality-related hidden features while eliminating redundancy and maintaining structural simplicity.
- The proposed approach offers superior effectiveness and performance for soft sensor applications in industrial processes.
