Theoretical Model Of Saturaetd Water Content In High Temperature And Pressure Sour Natural Gas

In the process of exploitation,gathering transportation and processing for hydrocarbon and high containg H₂S and CO₂ gas reservoirs,three preventions named blocking,corrosion and hydrate formation are the most important parts in the entire gas reservoir development,which are all closely related to saturated water content of natural gas.The saturated water content of natural gas is an important parameter during the natural gas production,processing and transportation,directly affecting the accuracy of calculation for the technology design of natural gas reservoirs.As a direct consequence,support vector regression and BP artificial neural network models combined with genetic algorithm,grey correlation analysis and outlier detection,were proposed to predict saturated water content of multiple systems,including hydrocarbon gases,pure H₂S and pure CO₂ gases,H₂S,CO₂ mixed gases and high H₂S,C02 acid natural gases.The proposed models were able to estimate saturated water content as a function of relative density of natural gas,CH4,CO₂,H₂S composition mole fraction and saturated water vapor equilibrium temperature and pressure.The suitable range of all the data of the models includes:temperature 0?325?,pressure 0?150MPa,H₂S mole fraction 0?100%,CO₂ mole fraction 0?100%.In the calculating process of GA-SVM and BP neural network models,based on the training of 316 experimental data points,the multi systems prediction of 83 data points for saturated water content were carried out.The results show that absolute average deviation of saturated water content for hydrocarbon gases,pure H₂S and pure CO₂ gases and high H₂S,CO₂ acid natural gases by using GA-SVM model is 1.38%,2.6%,4.3%and 6.28%,respectively,and the absolute average deviation for H₂S,CO₂ mixed gas is 11.93%by using BP neural network model.Compared with the commonly used ANN model,semi empirical model,Bukacek model,Awad model,AQUAlibrium software,simplified thermodynamic-Bahadori model,simplified thermodynamic-Mohammadi model,Bahadori-Khaled model,Wang Junqi model,modified thermodynamic model,the calculation results showed that GA-SVM and BP-ANN model can meet the requirements of Engineering precision because of high prediction accuracy,strong generalization ability and more stability.Finally,outlier diagnosis was performed on the basis of whole data sets to identify the applicable range of all models investigated in this work by detecting the probable doubtful points.The testing results show that the new GA-SVM and BP neural network models can be detected all data points with a minimum quantity of doubtful points and the best stability,which can provides a new comprehensive method for analyzing and predicting water content of hydrocarbon gases,pure H₂S and pure C02 gases,H₂S,CO₂ mixed gases and high H₂S,CO₂ acid natural gases and other systems.