Hybrid smart model to determine concentration of acidic gases in absorption tower of sweetening process

Abstract

AbstractThe natural gases produced from underground reserves may contain sour gases such as H2S and CO2. If the amount of these gases is greater than the standard level, it will be harmful to humans and the environment. Hence, it appears important to accurately determine the concentration of H2S and CO2 in various equipment/units of a natural gas sweetening plant. In this study, a new connectionist approach is introduced to obtain the concentration of outlet acid gases of the sweetening tower of the South Pars gas refinery (in Iran), by utilizing a conventional connectionist tool optimized by an imperialist competitive algorithm (ICA). This ICA strategy is applied to optimize the weights, biases, and number of neurons of the artificial neural network (ANN) model. The input parameters in this modelling strategy include time, the amount of H2S in the output amine, the flow rate of input sour gas, the temperature of sea water for cooling, the mass flow rate of low‐pressure steam to the amine–amine heat exchanger, and the flow rate of input amine. The performance of the hybrid deterministic tool, ANN‐ICA, is compared with the ANN‐back propagation (BP) method. The coefficient of determination and mean squared error to forecast the concentration of output H2S are 0.8931 and 0.0125 for the ANN‐BP algorithm and 0.9057 and 0.0104 for the ANN‐ICA algorithm, respectively. These statistical values are 0.8428 and 0.0001 for the ANN‐BP model and 0.9307 and 0.000 05 for the ANN‐ICA model, respectively, while predicting output CO2 concentration. The results confirm that the ANN‐ICA is a more reliable approach, compared to ANN‐BP, to forecast the concentration of acidic gases leaving the absorption tower.