Modeling solubility of sulfur in pure hydrogen sulfide and sour gas mixtures using rigorous machine learning methods

• Published in: International journal of hydrogen energy Vol. 45; no. 58; pp. 33274 - 33287
• Main Author: Nait Amar, Menad
• Format: Journal Article
• Language: English
• Published: Elsevier Ltd 27-11-2020
• Subjects: Cascaded forward neural network (CFNN); Gene expression programming (GEP); Multilayer perceptron (MLP); Natural gas; Sour gases; Sulfur solubility; Sour gases; Sulfur solubility; Multilayer perceptron (MLP); Gene expression programming (GEP); Natural gas; Cascaded forward neural network (CFNN)
• ISSN: 0360-3199; 1879-3487
• DOI: 10.1016/j.ijhydene.2020.09.145

Accurate determination of sulfur solubility in pure hydrogen sulfide (H2S) and sour gas mixtures has a leading role and a fundamental importance in handling and addressing sulfur deposition issues. In this study, rigorous paradigms based on two artificial neural network (ANN) types, namely multilayer perceptron (MLP) and cascaded forward neural network (CFNN), optimized by Levenberg–Marquardt (LM) algorithm were proposed as machine learning (ML) modeling tools to predict the solubility of sulfur in sour gas mixtures and pure H2S. Besides, explicit and simple-to-use correlations were established using gene expression programming (GEP). The paradigms derived from the methods aforementioned were developed using widespread experimental database. The obtained results indicated that the outcomes gained from the proposed MLP, CFNN and GEP-based correlations are in a high coherence and agreement with the experimental data. In addition, it was found that among the all suggested schemes, CFNN models are the most accurate paradigms for estimating the solubility of sulfur in sour gas mixtures and pure H2S with root mean square error (RMSE) of 0.0232 and 3.8101, respectively. Furthermore, a comparison between the performance of CFNN and the prior alternatives demonstrated that the CFNN models predict the solubility of sulfur in sour gas mixtures and pure H2S more accurately. Moreover, based on the trend analysis, it was concluded that the predictions of CFNN follow the real tendency of sulfur solubility in pure H2S and sour gas mixtures with respect to the input parameters. Besides, the sensitivity analysis dictated that pressure and temperature have the most significant impact on sulfur solubility calculation in pure H2S and sour gas mixtures. The results reported in this investigation revealed that implication of the considered soft computing approaches in the estimation of sulfur solubility in sour gas mixtures and pure H2S can lead to the generation of more reliable predictive paradigms which can be integrated in other related applications. Lastly, the findings of this study can help for effective prediction of the solubility of sulfur in sour gas mixtures and pure H2S while simulating different natural gas processes.
•Sulfur solubility in pure H2S and sour gases was modeled using MLP, CFNN and GEP.•239 experimental points were used for the development of the models.•The proposed paradigms and correlations generate excellent prediction performance.