Advanced machine learning computations for estimation of hydrogen solubility in oil samples: Model comparisons and validation

Abstract

For analyzing hydrogenation process for treatment of petroleum-based fuels, solubility of hydrogen in the feed should be well correlated. The main correlation factors are temperature and pressure which have a great effect on hydrogen solubility. This research paper presents the development of three models for predicting the solubility of hydrogen gas (H2) in diesel. Pressure and Temperature are the input parameters and solubility is single output. The models were fine-tuned using the Bat Algorithm (BA). The three models include Orthogonal Matching Pursuit Regression (OMP), K Nearest Neighbors Regression (KNN), and Tweedie Regression (TDR). The results of the study revealed that the OMP model achieved the highest level of accuracy, with an R2 score of 0.98, and the least RMSE and MAE error rates of 0.24 and 0.19, respectively. The KNN model also performed well with an R2 score of 0.92, an RMSE of 0.42, and an MAE of 0.37. The TDR model had the lowest accuracy compared to the other two models. These results imply that the OMP model is the most suitable one for predicting H2 solubility. The models can be used to enhance the efficiency of fuel production by providing accurate predictions of H2 solubility.

Keywords

• Petroleum refining
• Fuel treatment
• Diesel
• Hydrogen
• Modeling
• Computation