Prediction of natural gas hydrates formation using a combination of thermodynamic and neural network modeling

• Published in: Journal of petroleum science & engineering Vol. 182; p. 106270
• Main Authors: Rebai, Noura, Hadjadj, Ahmed, Benmounah, Abdelbaki, Berrouk, Abdallah S., Boualleg, Salim M.
• Format: Journal Article
• Language: English
• Published: Elsevier B.V 01.11.2019
• Subjects: Artificial neural network; Backpropagation algorithm; Gas hydrate formation; Van der Waals-Platteeuw models; Backpropagation algorithm; Artificial neural network; Gas hydrate formation; Van der Waals-Platteeuw models
• ISSN: 0920-4105; 1873-4715
• DOI: 10.1016/j.petrol.2019.106270

During the treatment or transport of natural gas, the presence of water, even in very small quantities, can trigger hydrates formation that causes plugging of gas lines and cryogenic exchangers and even irreversible damages to expansion valves, turbo expanders and other key equipment. Hence, the need for a timely control and monitoring of gas hydrate formation conditions is crucial. This work presents a two-legged approach that combines thermodynamics and artificial neural network modeling to enhance the accuracy with which hydrates formation conditions are predicted particularly for gas mixture systems. For the latter, Van der Waals-Platteeuw thermodynamic model proves very inaccurate. To improve the accuracy of its predictions, an additional corrective term has been approximated using a trained network of artificial neurons. The validation of this approach using a database of 4660 data points shows a significant decrease in the overall relative error on the pressure from around 23.75%–3.15%. The approach can be extended for more complicated systems and for the prediction of other thermodynamics properties related to the formation of hydrates. •Control and monitoring of gas hydrate formation conditions is crucial.•Van der Waals-Platteeuw thermodynamic model proves inaccurate for gas mixture systems.•An additional corrective term has been approximated to improve accuracy.•A trained Artificial Neuron network is deployed for that purpose.•A significant decrease in the overall relative error on the pressure is achieved.