Prediction Model Of Sulfur Solubility In Natural Gas With High H₂S Content

Natural gas is one of the most important components of China's energy structure.In recent years,a large number of sour gas reservoirs were successively developed.During the exploitation of sour gas,sulfur deposition is a serious problem in strata or pipelines due to the decrease of temperature and pressure.Sulfur deposition not only blocks formation pore,reduces formation permeability and seriously affects gas production,but also may cause serious engineering accidents.The existence of dissolved sulfur is the fundamental cause of sulfur deposition in sour gas.Study of the sulfur solubility in sour gas is the basis to solve the problem of sulfur deposition.The main research contents are as follows:?1?A prediction model of sulfur solubility in H₂S based on molecular dynamics is proposed and a micro-model is established according to gas-solid coexistence method.The average relative error between simulation results and experimental data is 11.05%.Compared with the three Chrastil-type prediction models,the prediction accuracy of the new model has been improved.?2?All experimental data of sulfur solubility in H₂S,CO₂ and CH₄ are used to evaluate the reliability of the experimental data.For the whole evaluated experimental data,it's found that 28%are considered as thermodynamically consistent,28%are inconsistent and 44%are deemed to be not fully consistent.All the experimental data of sulfur solubility in CH₄ are considered as unreliable.?3?A prediction model of sulfur solubility in sour gas based on machine learning is proposed.Support Vector Regression?SVR?is used to build the model and chaotic firefly algorithm?CFA?is used to optimize the parameters of SVR.Compared with the experimental data,the average relative errors of CFA-SVR of sulfur solubility in sour gas,H₂S and CO₂ are 4.51%,2.11%and 10.12%,respectively.The result obtained by using new CFA-SVR model is in good agreement with the experimental data,and the error is smaller than previous models based on semi-empirical formula or equation of state.