Data-driven measuring and hybrid modeling of racemic methionine crystallization thermodynamics in K2CO3 aqueous systems

Considering the intrinsic difficulty of accurately determining the thermodynamic data of complex impurity-containing systems in industrial crystallization processes, this study creatively establishes a predictive correlation between infrared spectra, operating conditions, and the thermodynamics of the racemic methionine ( dl -methionine) crystallization system in K 2 CO 3 aqueous systems. Using principal component analysis, random forest, and Gaussian process regression architecture, a concentration prediction machine learning sub-model was constructed, yielding a coefficient of determination of 0.988 in the test set. Subsequently, a first-principle sub-model combining equilibrium conditions with the electrolyte activity coefficient model was established by developing a parameter optimization program fitted experimental data to obtain a predicting error of 2.09%. The solubilization effect of carbonate salt on methionine was quantitatively characterized. The influence of temperature, pH, and carbonate component on the distribution and activity coefficients of dl -methionine species was thoroughly investigated. The hybrid model can better serve for the analysis of industrial dl -methionine crystallization processes.