[GSH fermentation process modeling using entropy-criterion based RBF neural network model].

Area of Science:

• Biochemical Engineering
• Computational Biology
• Machine Learning

Background:

• Fermentation process modeling, particularly for glutathione (GSH), faces challenges with prediction accuracy and generalization due to noisy experimental data.
• Traditional modeling approaches often rely on Mean Squared Error (MSE) criteria, which can lead to issues like over-learning and poor generalization.

Purpose of the Study:

• To develop a novel Radial Basis Function (RBF) neural network modeling approach for GSH fermentation that improves prediction accuracy and generalization.
• To address the limitations of existing methods by incorporating an entropy criterion for parameter learning.

Main Methods:

• A novel RBF neural network model was developed utilizing an entropy criterion for parameter learning.
• This entropy-based approach considers the entire distribution structure of the training data, unlike traditional MSE-based methods.
• The proposed model was applied to the specific case of GSH fermentation process modeling.

Main Results:

• The entropy-based RBF neural network demonstrated superior prediction accuracy compared to traditional methods.
• The model exhibited enhanced generalization capabilities, effectively avoiding over-learning issues.
• Improved robustness was observed in the modeling of the GSH fermentation process.

Conclusions:

• The proposed entropy-based RBF neural network offers a significant improvement for GSH fermentation process modeling.
• This method provides better prediction accuracy, generalization, and robustness, making it a valuable tool for biochemical process analysis.
• The approach holds potential for wider application in complex fermentation modeling scenarios.