Dynamic hybrid modeling of fuel ethanol fermentation process by integrating biomass concentration XGBoost model and kinetic parameter artificial neural network model into mechanism model

Abstract

Fuel ethanol has drawn extensive attention as renewable energy. However, modeling the fuel ethanol batch fermentation process is still a critical task. The unstructured kinetic model (UKM) is often utilized to model the process, but it encounters two problems due to the large differences in initial glucose concentrations. First, the UKM has poor predictions of the yeast growth, which is a crucial production index. Second, the kinetic parameters of the UKM are time-varying because of the changing environmental conditions. The constant manually set kinetic parameters affect the prediction accuracy. To tackle the problems, we propose a dynamic hybrid model of the fuel ethanol fermentation process. First, a biomass concentration prediction model based on extreme gradient boosting is developed. It predicts the values of biomass concentrations and mycelium growth rate as supplementary mechanism knowledge. Then, we present an artificial neural network-based model to determine the time-varying kinetic parameters. Our model can accurately predict the time series of biomass, ethanol, and glucose concentrations, with RMSEs reaching 0.3323, 1.9295, and 3.0540. Experimental results show that the dynamic hybrid model performs with satisfactory accuracy in modeling the fuel ethanol fermentation process.