Preference-driven yield-and-quality optimization for high-sulfur gas sweetening process by extreme learning machine model

Abstract

Yield and quality, as the two most important output variables of high-sulfur gas (HSG) sweetening process, are affected by the operating parameters. The HSG sweetening process involves more than ten operating parameters, so the relationship between the parameters and output variables is complex, non-linear and strong coupling. This paper tries to use data mining methods to explore this relationship and apply it to optimize the yield and quality. First, a ten (inputs)-to-three (outputs) model is established by extreme learning machine (ELM). Then, a preference-driven multi-objective optimization algorithm is used to maximize the yield while ensuring that the concentration of carbon dioxide (CO2) and the concentration of hydrogen sulfide (H2S) in the treated gas are close to but not exceeding 3% and 4 ppm respectively. The proposed method is validated in a HSG purification plant in southwest China. A set of 3044 production data is collected and randomly divided into 80 and 20% for training and testing. The results show that the established ELM model is in good agreement with the actual operation data. The maximum deviation of mean square error (MSE), mean absolute error (MAE) and average absolute deviation percent (AAD %) of the predictions in three scenarios are 0.2047, 0.3177 and 7.91% respectively. Moreover, the optimization based on the obtained ELM model is also validated. In particular, the H2S concentration and CO2 concentration in the treated gas are significantly higher than those before optimization, but have not exceeding the limits. Thus, the consumptions of energy and amine solvents decreased, while the yield increased.