An Accurate and Efficient Surrogate Model-Based Framework for Coupled Wellbore-Reservoir Modeling

Abstract In the development of oil and gas fields, accurately simulating the interactions between wellbores and reservoirs is crucial for enhancing the precision of production forecasts, optimizing development strategies, and aiding reservoir management decisions. Although significant research achievements have been made in the modeling of wellbore multiphase flow and reservoir permeation, and there are various mature commercial numerical simulation software available on the market, research on wellbore-reservoir coupling modeling still lags behind. The main challenges include mismatches in the temporal and spatial scales between wellbores and reservoirs, uncertainty in model parameters, and low computational resource efficiency, all of which severely impact the practicality and accuracy of coupling models. To address these issues, this study has developed a new integrated wellbore-reservoir coupling model using artificial intelligence technology. By constructing separate surrogate models for the wellbore and reservoir—the wellbore surrogate model employs neural differential equations, and the reservoir surrogate model uses an autoencoder for dimensionality reduction followed by prediction with neural differential equations—this study effectively achieves high-fidelity and high-efficiency in model coupling. The integration of these two surrogate models not only resolves the main challenges faced by traditional coupling models but also facilitates model integration, ultimately creating a high-fidelity, high-efficiency, and multi-scale adaptable coupling model, providing a powerful tool and a solid theoretical foundation for achieving integrated simulation of wellbores and reservoirs.