Title: Prediction of pressure gradient in gas liquid flow based on meta-learning methods

Authors: Zhenzhen Dong; Zhanrong Yang; Weirong Li; Xiaowei Zhang; Wei Guo; Tong Hou; Guoqing Dong

Addresses: Xi'an Shiyou University, Xi'an, 710065, China ' Xi'an Shiyou University, Xi'an, 710065, China ' Xi'an Shiyou University, Xi'an, 710065, China ' Research Institute of Petroleum Exploration and Development, Beijing, 100083, China ' Research Institute of Petroleum Exploration and Development, Beijing, 100083, China ' Xi'an Shiyou University, Xi'an, 710065, China ' Xi'an Shiyou University, Xi'an, 710065, China

Abstract: Gas liquid flow models in wellbores are crucial across various stages of the petroleum industry. Accurate pressure drop prediction in these flows is vital for optimal production schedules. Traditional models face challenges such as uncertain prediction boundaries and complex parameter computations, lacking universal applicability. Leveraging advances in machine learning, this paper introduces a predictive model for pressure gradients in gas liquid flow. Utilising 862 samples from experimental literature, the data were preprocessed, cleaned, and split into training and validation sets. Employing R² and RMSE as evaluation metrics, the optimal model was identified using meta-learning, combining decision tree, random forest, K-nearest neighbours, XGBoost and LightGBM as base models, with XGBoost as the meta-model. The model achieved an R² of 0.9826 and RMSE of 2.8200, highlighting key factors influencing gas liquid flow and emphasising the model's accuracy and significance.

Keywords: pressure gradient; gas liquid flow; machine learning; meta-learning.

DOI: 10.1504/PCFD.2025.149928

Progress in Computational Fluid Dynamics, An International Journal, 2025 Vol.25 No.6, pp.352 - 369

Received: 29 Aug 2024
Accepted: 10 Apr 2025
Published online: 18 Nov 2025 *

Full-text access for editors Full-text access for subscribers Purchase this article Comment on this article