A Novel CatBoost-Optuna Model for Hydrate Phase Equilibrium Prediction Considering the Molecular Properties and Functional Group Features

Summary The interaction mechanism among the components of each phase in systems containing organic solutes is highly complex, making it extremely difficult to precisely model the hydrate phase equilibrium based on the classic thermodynamics theory. To address this issue, an intelligent prediction model for hydrate phase equilibrium is developed, considering the molecular characteristics and functional group properties of organic substances. The main factors considered in the model are (i) development of a comprehensive experimental database covering various types of organic substances, component concentrations, and temperature-pressure ranges, in which the data imbalance problem is addressed using the Boruta feature selection and oversampling algorithms; (ii) feature extraction and optimization via integrating molecular characteristics and functional group information to enhance the model accuracy and universality; and (iii) an intelligent model for hydrate phase equilibrium prediction incorporating the CatBoost model and Optuna framework, which is of high generalization ability in handling nonlinear data and efficiency in hyperparameter optimization. The experimental results show that the proposed CatBoost-Optuna model performs the best among various machine learning models, with accuracy ranking from high to low as CatBoost-Optuna, CatBoost, extreme gradient boosting (XGBoost), random forest (RF), decision tree (DT), linear regression (LR), and support vector machine (SVM). Compared with traditional thermodynamic models, the performance of the developed model is significantly improved, with the coefficient of determination (R²) increasing from 0.856 to 0.990. By using the Optuna framework for hyperparameter optimization, the R2 of the model increases from 0.964 to 0.991. Furthermore, the developed model demonstrates excellent generalization ability in the validation of extrapolated data sets and quantitatively evaluates the influence of different organic functional groups, which can provide support for the design of hydrate formation/inhibitors.