Abstract
Frequently, petroleum refineries create a variety of fuels as well as a vast range of chemicals for diverse applications. One of the most frequent procedures for purifying petroleum products from unwanted sulfur species and reducing SO2 emissions is the hydrodesulfurization (HDS) process. However, HDS is still challenging since a variety of factors influence sulfur removal rates, including operating circumstances, feed compositions, catalyst activity, and so on. In actuality, reducing sulfur compounds comes at a high price, both environmentally and economically. In practice, it is necessary to forecast process yields and their implications for productivity, profitability, and environmental considerations. The study of such outcomes could serve as guidance for scholars and practitioners alike. Machine Learning (ML) algorithms have proven to be effective in solving various real-world problems in engineering and industrial fields, including the petroleum industry. This study presents a four-input support vector regression (SVR) model hybridized with a Bayesian optimization to predict three yields of the HDS process including outlet sulfur concentration, percentage of SO2 emission, and percentage of biphenyl. The proposed models are used to identify the best laboratory configuration for better optimization of the HDS process. The obtained modeling results reveal that the proposed models are competent with a high degree of accuracy. The correlation coefficients during the testing of the three models were 99.1, 99.2, and 98.8% while the average experimental errors RMSE and MRAE were 0.022 and 0.097, respectively.