Optimal modeling of fermentation process using hybrid support vector regression

This study investigates the accuracy of concentration prediction in industrial fermentation, a critical factor for optimizing large-scale production. To address the limitations of single models in generalization, a hybrid support vector regression (H-SVR) model is proposed, combining the strengths of flexible and robust SVRs to enhance prediction accuracy. The algorithm segments fermentation data based on bacterial characteristics at different stages, emphasizing local phase-specific features, and uses weighted factors to construct the final hybrid model. The corresponding hyperparameters are optimized via grid search to ensure performance. Simulation results based on industrial penicillin fermentation data and succinic acid fermentation experiment demonstrate that the H-SVR model significantly reduces prediction error compared to models such as least squares support vector machines and some network models, while enabling real-time process monitoring. These findings highlight the potential of the H-SVR model in complex biological systems and demonstrate its effectiveness as a tool for optimizing fermentation processes.