Multiobjectives optimization in petroleum refinery catalytic desulfurization using Machine learning approach

Abstract

In the petroleum refining industries, the hydrodesulfurization (HDS) process is one of the most common techniques for purifying petroleum products from undesirable sulfur species to reduce SO2 emission. However, HDS is still complicated as many aspects dominate the rate of sulfur removal such as operating conditions, feed formulations, catalyst activity, etc. In reality, the reduction of sulfur compounds has expensive costs either environmentally or economically. In practical usage, there is a need to predict the process yields and their consequences related to productivity, profitability, and environmental factors. The study of such consequences may serve as a guide for researchers and practitioners. Artificial Intelligence algorithms have proven to be effective in solving various real-world problems in engineering and industrial fields, including the petroleum industry. In this paper, a hybrid-ML approach (SVMG) incorporates both support vector machine (SVM) and genetic algorithm (GA) to build models for HDS yields prediction. The created SVMG models are used to identify the best laboratory configuration for better optimization of the process. The optimization problem can be stated as minimization of sulfur concentration, percentage of SO2 emission, and HDS cost. The obtained modeling results reveal that the developed SVMG models are competent with a high degree of accuracy as indicated by correlation coefficients of 99.4%, 99.0%, and 98.7% during the testing of the models. In addition, the results of the experimental validation showed that there is an agreement between the expected values obtained from the proposed models and experimental values that were conducted in the laboratory with an average experimental error of less than 4%.