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

- A new variable selection method for a soft-sensor was developed.
- The proposed variable selection methods are based on the correlation between variables and group Lasso.
- Parameter tuning in the proposed method is easier than conventional methods.
- The proposed method was applied to soft-sensor design of a pharmaceutical process and a chemical process.
- The proposed method could reduce the number of input variables of a soft-sensor and improve its performance.

Appropriate input variables have to be selected for building highly accurate soft sensor. A novel input variable selection method based on nearest correlation spectral clustering (NCSC) has been proposed, and it is referred to as NCSC-based variable selection (NCSC-VS). Although NCSC-VS can select appropriate input variables, a lot of parameters have to be tuned carefully for selecting proper variables. The present work proposes a new methodology for efficient input variable selection by integrating NCSC and group Lasso. The proposed NCSC-based group Lasso (NCSC-GL) can not only reduce the number of tuning parameters but also achieve almost the same performance as NCSC-VS. The usefulness of the proposed NCSC-GL is demonstrated through applications to soft sensor design for a pharmaceutical process and a chemical process.