Prediction of the sulfur solubility in pure H2S and sour gas by intelligent models

Abstract

In order to reduce time and enhance accuracy, four intelligent models named Grey Wolf Optimizer based Least Squares Support Vector Machine (GWO-LSSVM), Grey Wolf Optimizer based Radial Basis Function (GWO-RBF), Genetic Algorithm based Adaptive Network Fuzzy Inference System (GA-ANFIS), and Particle Swarm Optimization based Adaptive Network Fuzzy Inference System (PSO-ANFIS) were applied to predict sulfur solubility in pure H2S and sour gas. According to Pearson correlation analysis, the content of H2S, critical temperature, temperature, gas density, and pressure were selected as input variables and sulfur solubility was selected as an output variable in sour gas. The contradistinction among the four models reveals GWO-LSSVM behaves the best performance with the minimum average absolute relative deviation (AARD = 3.5029%), and the maximum determination coefficient (R2 = 0.9976) in all 239 data. But according to the minimum root mean squared error (RMSE), PSO-ANFIS performs best in pure H2S and sour gas among the four models. The leverage method was used to search outlier data for sulfur solubility, indicating that there are only 5 anomalous data points of all 239 data for the best GWO-LSSVM model.