Soft sensing modeling of penicillin fermentation process based on local selection ensemble learning.

Area of Science:

• Chemical Engineering
• Bioprocess Engineering
• Machine Learning

Background:

• Penicillin fermentation exhibits complex nonlinear relationships between input and output variables.
• Existing soft sensing models struggle to meet the stringent accuracy demands of chemical production.
• Accurate prediction is crucial for optimizing penicillin fermentation processes.

Purpose of the Study:

• To develop an advanced soft sensing modeling strategy for penicillin fermentation.
• To enhance prediction accuracy in nonlinear, multi-output systems.
• To address the limitations of current modeling approaches in industrial bioprocesses.

Main Methods:

• A novel localization method using transfer entropy and k-means clustering to reconstruct sample sets.
• Multi-objective support vector regression (SVR) for establishing local soft sensing models.
• Selective ensemble learning with adaptive weight calculation for sub-models.
• Sparrow Search Algorithm (SSA) for optimizing model parameters and mitigating adverse effects.

Main Results:

• The proposed local selective ensemble learning multi-objective soft sensing strategy demonstrates superior prediction performance.
• The method effectively handles the strong nonlinear relationships inherent in penicillin fermentation.
• Compared to existing techniques, the new strategy achieves higher accuracy and reliability.

Conclusions:

• The developed modeling strategy offers a significant improvement for soft sensing in penicillin fermentation.
• The integration of localization, ensemble learning, and advanced optimization techniques enhances predictive capabilities.
• This approach provides a robust solution for meeting the accuracy requirements in chemical production.