Analysis of influencing factors and prediction of quality concentration in penicillin fermentation process

Abstract

To address process optimization challenges from nonlinear dynamics and strong multi-parameter coupling in penicillin fermentation, this study proposes a method integrating statistical analysis and machine learning for factor analysis and concentration prediction. First, the non-parametric statistical methods was used to detect differences in 7 influencing factors across different penicillin concentration groups, and the RF algorithm was applied to calculate their weights. Subsequently, XGBoostR, RFR and SVR models for penicillin concentration prediction were established, with parameters optimized via Bayesian optimization. Experiments on a simulation dataset showed that the RFR model performed optimally with high accuracy and robustness, Its R2, RMSE and MAE are 0.9755, 0.1480, and 0.0875, respectively. The method constructed in this study provides scientific reference for process parameter control and quality monitoring in penicillin fermentation.