Natural
gas hydrates have become a threat to natural gas companies
and oil industries for the formation of gas hydrates in transfer pipelines
can lead to pipe blockage. Natural gas hydrate formation is favored
by low temperature and high pressure; thus, if the right combination
of temperature and pressure are well investigated, it is possible
to solve the pipe blockage problems. Therefore, developing a precise,
easy-to-use method to predict natural gas hydrate formation temperature
(HFT) is very important. In this paper, we are intended to develop
a novel machine learning model called twin support vector regression
(TSVR) for predicting HFT in a wide range of natural gas mixtures.
For constructing the TSVR model, 513 experimental data in the system
methane, ethane, propane, butane, pentane, nitrogen, carbon dioxide/water,
methane, carbon dioxide, propane, methanol, sodium chloride, and calcium
chloride have been collected from open literature. We compared the
performance of the TSVR model with least squares support vector regression
(LSSVR) and three widely used correlations in the system methane,
ethane, propane, butane, pentane, nitrogen, and carbon dioxide, which
show that the TSVR model has fewer deviations than the LSSVR model
and three correlations. The values of root-mean-square error (RMSE),
mean absolute percentage error (MAPE), and correlation factor (R
2) are 2.4010, 0.0061, and 0.9108, respectively.
In addition, statistical parameters show that the TSVR model can also
surpass the LSSVR model in predicting HFT in the system water, methane,
carbon dioxide, propane, methanol, sodium chloride, and calcium chloride.