Supercritical water gasification thermodynamic study and hybrid modeling of machine learning with the ideal gas model: Application to gasification of microalgae biomass

This study presents a hybrid modeling approach that combines a simplified phenomenological model with machine learning techniques for predicting variables in the microalgae biomass gasification process in supercritical water (SCWG). The simplified phenomenological model, based on the Gibbs energy minimization methodology (minG) associated with the ideal gas model, exhibits significant deviations when compared to the actual behavior of the process. A machine learning model is employed as a corrective tool to enhance prediction accuracy and mitigate errors inherent in the ideal gas model. The simplified model (minG + Ideal Gas) yields moderate predictions for hydrogen formation (H2MinG + Ideal Gas) throughout the microalgal biomass SCWG process (r2 < 0.92). To develop an easily implementable tool, a linear regression model is used to predict the discrepancies between the simplified approach and the real process results (ΔH2), resulting in a good fit (r2 > 0.99). The deviations predicted by the machine learning model are incorporated into the results generated by the simplified phenomenological model, producing the prediction of the hybrid model (H2MinG + Ideal Gas+ΔH2), which shows a good fit with the actual hydrogen formation data (r2 > 0.99).