Answers are grounded in this paper and cite their sources.
A multiphase flow simulator and curated datasets are shared with the community to facilitate research on machine learning and other data-driven methodologies. The simulator is based on a drift-flux model and provides low-fidelity, one-dimensional, steady-state solutions that capture the main characteristics of multiphase flow in vertical wells. A sampling strategy for the model parameters is proposed which results in a diverse set of oil and gas wells with different geometric, fluid, and thermal properties. Sampling strategies are also devised to simulate stationary and non-stationary boundary conditions and different operating modes. The method is used to generate three large datasets, each with one million data points from 2000 wells. The simulated data is compared to real production data from more than 300 wells in a comprehensive data analysis. The simulator implementation is validated by demonstrating accurate predictions of pressures and flow rates for a real well. Finally, the generated datasets are applied in a couple of machine learning examples.
No takes yet. Share an insight, caveat, or question.
Grimstad et al. (2025) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: