Efficient input variable selection for soft-senor design based on nearest correlation spectral clustering and group...

Area of Science:

• Process Engineering
• Data Science
• Chemometrics

Background:

• Accurate soft sensor models require careful input variable selection.
• Existing methods like nearest correlation spectral clustering (NCSC) are effective but parameter-intensive.

Purpose of the Study:

• To develop an efficient input variable selection methodology for soft sensors.
• To reduce the number of tuning parameters in soft sensor design.
• To maintain or improve the performance of soft sensor models.

Main Methods:

• Integration of nearest correlation spectral clustering (NCSC) with group Lasso.
• Development of the NCSC-based group Lasso (NCSC-GL) method.
• Application of NCSC-GL to soft sensor design for pharmaceutical and chemical processes.

Main Results:

• The proposed NCSC-GL method significantly reduces the number of tuning parameters.
• NCSC-GL achieves performance comparable to the more complex NCSC-VS method.
• The method demonstrates practical utility in real-world process applications.

Conclusions:

• NCSC-GL offers an efficient and effective approach for input variable selection in soft sensor development.
• This methodology streamlines the soft sensor design process, particularly in complex industrial settings.
• The integration of NCSC and group Lasso provides a robust solution for optimizing soft sensor accuracy and usability.