Recurrent Neural Networks-Based Model Predictive Control for Continuous Fermentation Process in 1G/2G Ethanol Production Plant

This study presents the application of a Recurrent Neural Network-based Model Predictive Control (RNN-MPC) strategy for optimizing a continuous fermentation process. The RNN model, trained with a Mean Squared Error (MSE) of 7.14× 10 ^-5 , accurately predicted process dynamics and was integrated into the MPC framework. Optimal control parameters, including a prediction horizon of 10 hours and a control horizon of 1 hour, were identified to balance system stability with computational efficiency. Comparative analysis demonstrated that the RNN-MPC outperformed traditional PID control, achieving faster settling times (20 hours vs. 90 hours), reduced overshoot (0.2 vs. 2–5), and enhanced robustness to disturbances. These results underscore the effectiveness of RNN-MPC in managing complex bioprocesses and highlight its potential for broader industrial application.