A Hybrid Approach for Steady-State Production Optimization of a Real Oil and Gas Platform: Integrating Physics-Based Models, Machine Learning Techniques, and Field Monitoring …
Trevisan, Luca Shokry, Ahmed Montini, Marco Moulines, Eric Zio, Enrico
Trevisan, Luca
Shokry, Ahmed
Montini, Marco
Moulines, Eric
Zio, Enrico
Supervisor
Department
Machine Learning
Embargo End Date
Type
Journal article
Date
2025
License
Language
English
Collections
Research Projects
Organizational Units
Journal Issue
Abstract
This work concerns the optimization of the daily production of a real offshore oil and gas network. The lack of accurate physics-based models for fundamental units in the system, particularly the multiphase pumps, makes it unfeasible to obtain an accurate physics-based model of the entire network to be used for optimization purposes. Therefore, this work innovatively addresses this issue by developing a hybrid model for the real oil and gas network under study. Extensive monitoring data obtained from the real multiphase pumps over an extended timeframe form the foundation for developing machine learning (ML) models, which are adept at capturing the complex relationships within the pump system. Subsequently, we integrate these ML-based pump models with the existing physics-based models of other network components, including wells, gathering networks, risers and separators. These physics-based models are crafted using state-of-the-art industrial software, ensuring robustness and accuracy in representing the components’ actual behavior. The hybrid model's predictive capabilities are validated against real data from the offshore network, affirming its ability to accurately capture system behavior. Leveraging this validated hybrid model, optimization is performed using a differential evolution algorithm, for maximizing production efficiency while adhering to operational constraints. Our outcomes underscore several key findings: firstly, the ML-based pump models demonstrate remarkable accuracy in approximating the intricate relationships among pump variables, secondly, the hybrid model exhibits commendable predictive accuracy, effectively simulating the real behavior of the whole offshore production network; finally, the optimization yields tangible production enhancements, surpassing the network’s actual performance under historical operating conditions.
Citation
L. Trevisan, A. Shokry, M. Montini, E. Moulines, and E. Zio, “A hybrid approach for steady-state production optimization of a real oil and gas platform: Integrating physics-based models, machine learning techniques, and field monitoring signals,” Comput Chem Eng, vol. 203, p. 109320, Dec. 2025, doi: 10.1016/J.COMPCHEMENG.2025.109320
Source
Computers & Chemical Engineering
Conference
Keywords
Oil and Gas, Subsea Networks, Real-Time Production Optimization, Surrogate-Based Optimization, Hybrid Models, Machine Learning, Artificial Neural Network, Random Forest Regression
Subjects
Source
Publisher
Elsevier