Prediction of Temperature-Dependent\nHenry’s\nLaw Constants by Matrix Completion

Methods for predicting Henry’s law constants <i>H</i><sub><i>ij</i></sub> describing the solubility\nof solutes <i>i</i> in solvents <i>j</i> as a\nfunction of temperature\nare essential in chemical engineering. While isothermal properties\nof binary mixtures can conveniently be predicted with matrix completion\nmethods (MCMs) from machine learning, we advance their application\nto the temperature-dependent prediction of <i>H</i><sub><i>ij</i></sub> in the present work by combining them with\nphysical equations describing the temperature dependence. For training\nthe methods, experimental <i>H</i><sub><i>ij</i></sub> data for 122 solutes and 399 solvents ranging from 173.15\nto 573.15 K were taken from the Dortmund Data Bank. Two MCMs are proposed:\na data-driven MCM that relies solely on experimental data and a hybrid\nMCM that incorporates predictions from the established Predictive\nSoave-Redlich-Kwong (PSRK) equation of state (EoS), effectively combining\nphysical knowledge and machine learning. The performance of these\nMCMs is assessed via leave-one-out analysis and compared to that of\nthe PSRK-EoS, demonstrating superior prediction accuracy.