基于支持向量机的酸性天然气水合物生成条件预测

Prediction of hydrate formation condition of sour natural gases by using support vector machine(SVM)

针对高含H2S、CO2天然气水合物生成条件预测准确性较低的问题,引入1种酸性气体（H2S＋CO2）相对于CH4的贡献因子（σ）,建立一种基于支持向量机（support vector machine,SVM）预测高含H2S、CO2天然气水合物生成条件的新模型。该模型以σ、CH4、H2S、CO2摩尔分数和水合物生成温度为输入变量,以水合物生成压力为输出变量。在SVM模型59个实验点训练的基础上,进行了20个数据点的水合物生成条件预测。最后,基于杠杆方法,开展了所有数据（79个）的异常点检测。结果表明：在n（H2S）=4.95%~26.62%、n（CO2）=6.81%~22.30%条件下,SVM模型预测的平均绝对相对偏差（average absolute relative deviation,AARD）为5.70%,且所有数据点均未出现异常值。从而为酸性天然气水合物生成条件的准确预测提供了1种行之有效的新方法。

To overcome the low accuracy in predicting hydrate formation conditions of high H2S/CO2-content sour natural gases,a contribution factor（σ）was introduced to support vector machine（SVM）to predict hydrate formation conditions.The contribution factor（σ）,mole fraction of respectively CH4,CO2 and H2S,and hydrate formation temperature have been set as input variables and hydrate formation pressure as an output variable.59 out of 79 data points have been trained to search for the best SVM model structure,and the remaining 20 data sets have been used to check the capability of generalization for the best trained SVM model.Moreover,on the basis of Leverage mathematical approach,outlier diagnostics were performed to identify the range of applicability of the proposed SVM model and quality of existing experimental data.Results show that the predictions of the proposed model are in excellent agreement with experimental data accessible to published literature over the ranges of H2S=4.95%~26.62% and CO2=6.81%~22.30% with the average absolute relative deviation（AARD）of 5.70%,and that the presented SVM model for representation of hydration formation conditions of sour natural gases with high H2S/CO2 content is statistically valid and correct with no probable doubtful data points.These results provide an effective method to accurate prediction of hydrate formation conditions for sour natural gases.