Numerical–Machine Learning Hybrid Framework for Forecasting Multi-Phase Flow Dynamics in Deep Wellbores

Raghad S. Shamsah, Petroleum Engineering Department, College of Engineering, University of Kerbala, Karbala 56001, IraqFollow
Farhan Lafta Rashid, Petroleum Engineering Department, College of Engineering, University of Kerbala, Karbala 56001, IraqFollow
Mudhar A. Al-Obaidi, Technical Instructor Training Institute, Middle Technical University, Baghdad 10074, IraqFollow
Ahmed N. Al-Dujaili, Amirkabir University of Technology/ Petroleum Engineering Department,, No. 350, Hafez Ave, Valiasr Square, Tehran, Iran 1591634311Follow

Abstract

The Conventional models do not usually reflect the intricate relationships between the fluid characteristics, wellbore geometry, and operating flow parameters in the deep wellbores, which becomes a major issue in the extraction of hydrocarbons, where the unexpected behavior of the flows (bubble, slug, annular) can cause the efficiency and integrity of the equipment to be compromised. This problem has been tackled by developing a new neural network-based method that incorporates mathematical models and machine learning in this paper to identify the flow patterns and optimize the production parameters. It is an experimental study, where 12 input variables are used (e.g., depth, pressure, viscosity, and inclination of well-bore), which is processed by a trained neural network (MATLAB-based) utilizing experimental simulations based on a lab-scale wellbore setup with pressure, temperature, and flow sensors. Two models were experimented by neural network, first model is found to have 100% validation accuracy in predicting slug flow in cases of enrichment with complete input features, which is half the case with simple parameters, indicates that the four fundamental parameters cannot offer enough discriminative power measures among flow patterns, irrespective of network training. The most important results are found in: (A) determination of annular flow dominance at deeper levels, which is associated with steep pressure gradients (stability critical); (B) optimized production rates (oil: 0.052 m³/s, gas: 0.050 m³/s) with high stability (indicator: 0.76); and (C) high predictive capability, as indicated by error histograms around zero and correlation plots with R-values exceeding 0.99. The statistical significance (p < 0.05) in ANOVA provided by the model is another confirmation that it is a robust model.

Article Type

Original Study

First Page

Last Page

Creative Commons License

This work is licensed under a Creative Commons Attribution-NonCommercial 4.0 International License

Recommended Citation

Shamsah, Raghad S.; Rashid, Farhan Lafta; Al-Obaidi, Mudhar A.; and Al-Dujaili, Ahmed N. (2026) "Numerical–Machine Learning Hybrid Framework for Forecasting Multi-Phase Flow Dynamics in Deep Wellbores," Warith Scientific Journal of Engineering and Technology: Vol. 2: Iss. 1, Article 3.
DOI: https://doi.org/10.65892/3079-0697.1005

Download

COinS