Reliable estimation of optimal sulfinol concentration in gas treatment unit via novel stabilized MLP and regularization network

Abstract

Proper technology or configuration for sulfur recovery units (SRUs) strongly depends on H2S concentration of inlet acid gas stream. Various acid gas enrichment (AGE) schemes with different solvents can be used to reduce the concentration of carbon dioxide and heavy aromatic hydrocarbons while enriching the H2S content of SRU feed stream. The present article uses combinations of Aspen-HYSYS software and two in-house artificial neural networks (namely, Regularization and stabilized multilayer perceptron networks) to compare the AGE capability of sulfinol-M (sulfolane + MDEA) solvent at optimal concentration with traditional MDEA solution when both of them are used in a conventional gas treating unit (GTU). The simulation results indicate that the optimal concentration of Sulfinol-M aqueous solution (containing 37 wt% Sulfolane and 45 wt% MDEA) will completely eliminate toluene and ethylbenzene from the SRU feed stream while removing 80% of benzene entering the GTU process. Furthermore, mole fraction of H2S in the SRU feed stream increases from the conventional 33.48 mol % to over 57 mol %. Increased H2S selectivity of optimal sulfinol-M aqueous solution will elevate the CO2 slippage through sweet gas stream at around 4.5 mol % which is still below the permissible threshold. To the best of our knowledge, the stabilized MLP network has not been addressed previously.