A dynamic soft senor modeling method based on MW-ELWPLS in marine alkaline protease fermentation process.

Area of Science:

• Biochemical Engineering
• Process Control
• Machine Learning Applications

Background:

• Real-time online measurement of vital state variables in marine alkaline protease (MP) fermentation is challenging.
• Optimal control of MP fermentation is hindered by the lack of accurate real-time data.
• Existing methods struggle to provide reliable online monitoring for complex fermentation processes.

Purpose of the Study:

• To develop a dynamic soft sensor modeling method for accurate real-time prediction of vital state variables in MP fermentation.
• To improve the optimal control strategies for MP fermentation processes.
• To enhance the overall efficiency and yield of marine alkaline protease production.

Main Methods:

• A novel dynamic soft sensor model combining Just-in-Time Learning (JITL) and ensemble learning.
• Utilized Local Weighted Partial Least Squares (LWPLS) with a JITL strategy as the foundational modeling approach.
• Employed Moving Window (MW) for sub-dataset division and diversity selection, followed by stacking ensemble learning to fuse MW-LWPLS sub-models.

Main Results:

• The proposed JITL-ensemble learning method demonstrated superior prediction accuracy for vital state variables in MP fermentation.
• Experimental and simulation results confirmed the effectiveness of the MW-LWPLS sub-model fusion strategy.
• The developed soft sensor significantly outperformed traditional methods in predicting key fermentation parameters.

Conclusions:

• The dynamic soft sensor model effectively addresses the challenge of real-time online monitoring in MP fermentation.
• The combination of JITL, LWPLS, and ensemble learning provides a robust and accurate prediction tool.
• This approach offers a promising solution for optimizing and controlling marine alkaline protease fermentation processes.