Enhancing Wellbore Transient Multiphase Flow Simulation with a Surrogate Model Utilizing Neural Differential Equations

Abstract Transient multiphase flow in wellbores is crucial for the production of oil and gas wells, impacting key areas such as gas well liquid loading prediction, hydrate development, as well as safe operation and risk management. Currently, traditional wellbore flow simulation relies heavily on commercial software like OLGA, which, although powerful, is predominantly based on numerical methods, thus resulting in high computational costs and slow response times. In the context of the rapid development of digital twin technology, this mode of simulation can no longer meet the needs for real-time data processing and swift decision-making. Moreover, as oil and gas field development increasingly moves towards integration, coupling wellbores with reservoirs becomes particularly necessary. However, traditional numerical simulation models struggle to address the mismatch in temporal and spatial scales between wellbores and reservoirs. Therefore, developing a new generation of high-fidelity, efficient models is crucial. At the same time, surrogate models based on machine learning techniques have shown significant research interest and practical value in fields such as computational fluid dynamics (CFD) and medical imaging, providing a viable research direction. Nevertheless, the application of such models in petroleum engineering, especially in wellbore flow simulation, is still in its early stages. This study introduces a surrogate model for transient multiphase flow in wellbores based on neural differential equations. Preliminary testing has shown that this model has high computational efficiency and accuracy, effectively supporting the application of big data onsite and facilitating rapid decision-making. Additionally, the model employs a rolling prediction method with the capability of adaptive time stepping, which significantly addresses the mismatch in time scales between wellbores and reservoirs, thus offering the potential for high-precision and efficient coupling of wellbore-reservoir systems. Although the model still requires further improvements, it has already demonstrated potential and broad application prospects in the simulation of wellbore transient flows. Future work will focus on optimizing and expanding the application of the model to further enhance its usability and effectiveness in actual oil and gas production.