Evaluation of influential parameters for supersonic dehydration of natural gas: Machine learning approach

Abstract

The supersonic dehydration of natural gas is gaining more attention due to its numerous advantages over the conventional natural gas dehydration technologies. However, supersonic separators have seen minimal field applications despite the multiple benefits over other gas dehydration techniques. This has been mostly attributed to the uncertainty in ascertaining the design and operating parameters that should be monitored to ensure optimum dehydration of the supersonic separation device. In this study, the decision tree machine learning model is employed in investigating the effects of design and operating parameters (inlet and outlet pressures, nozzle length, throat diameter, and pressure loss ratio) on the supersonic separator performance during dehydration of natural gas. The model results show that the significant parameters influencing the shock wave location are the pressure loss ratio and nozzle length. The former was found to have the most significant effect on the dew point depression. The dehydration efficiency is mainly dependent on the pressure loss ratio, nozzle throat diameter, and the nozzle length. Comparing the machine learning model-accuracy with a 1-D iterative model, the machine learning model outperformed the 1-D iterative model with a lower mean average percentage error (MAPE) of 5.98 relative to 15.44 as obtained for the 1-D model.

Keywords

• Supersonic separator
• Machine learning model
• Natural gas dehydration
• Separation efficiency