Modeling and optimization of hydrogenation for crude oil by estimating hydrogen solubility in the solvent at...

A novel approach was introduced to predict the solubility of hydrogen in Athabasca bitumen sample by leveraging a hybrid approach based on Harmony Search Algorithm (HS) and AdaBoost. The solubility of H₂ in the samples is of great importance for treatment of heavy hydrocarbon in petroleum processing and can help optimize processes. In the correlation of data, pressure and temperature were used as the inputs, while the hydrogen solubility was assigned the sole response for the modeling. This approach was applied to three popular regression models: K-Nearest Neighbors (KNN), Theil-Sen, and Lasso, resulting in hybrid models named HSA-KNN, HSA-TS, and HSA-LAS, respectively. The HS algorithm is used to optimize the hyperparameters of the Adaboost and base models, and then AdaBoost is applied to enhance the performance of the base models. The HSA-KNN model achieved an R² score of 0.96466, MSE of 6.2790E-03, and a maximum error of 1.80485E-01, while the HSA-TS model achieved an R² score of 0.96433, MSE of 6.3994E-03, and a maximum error of 1.40763E-01. The HSA-LAS model, on the other hand, achieved an R² score of 0.89249.