Transparent predictive modelling of catalytic hydrodesulfurization using an interval type-2 fuzzy logic

Optimizing the desulfurization process in reactors would assist in predicting the performance of catalysts, and in turn, the final sulfur concentration in fuel. Machine learning (ML) techniques have proven their predictive capacity in solving challenging problems in the petroleum industry. Several ML-based sulfur predictors have been designed to predict the sulfur concentration in fuel products. The capability to handle the impression and uncertainty present in the real-world environment has made fuzzy logic (FL) one of the most commonly used soft computing paradigms in modeling. This paper proposes an interval type-2 fuzzy logic model to optimize the use of catalysts for hydrodesulfurization (HDS) in an oil refinery for cleaner production. Such optimization assists decision makers in identifying the precise required conditions such as temperature, pressure and the amount of catalyst required to avoid unexpected redesulfurization. The catalyst was synthesized from AlMoCo modified with boron, phosphorus or bismuth. A series of hydrodesulfurization was performed using these catalysts under various conditions. The collected data were used to build and test the model. The results showed a promising prediction performance in terms of average absolute relative error (AARE = 0.0647) and the squared correlation coefficient (R²=0.995). Further experimental validation done on unexplored parameter settings demonstrates that the predicted values obtained from the proposed model corresponds closely with the follow-up experimental results where an average absolute difference of less than 4 ppm was recorded. This proves the capacity of the interval type-2 FL model in handling uncertainty, which holds great promise for cleaner production of oil. The reported model prove to be good for prediction of catalysts performance in a single unit, however, it cannot be used for catalysts in dual units.