Stone Ridge Enhances Reservoir Simulations with NVIDIA Modulus on AWS

Stone Ridge Technology has made significant strides in the realm of reservoir simulation by integrating NVIDIA Modulus on AWS, according to a recent report by NVIDIA. The collaboration aims to enhance the efficiency and accuracy of reservoir simulations, which are crucial for addressing the inherent uncertainties in energy exploration.

Challenges in Reservoir Simulation

Energy exploration is fraught with uncertainties, including unknown geological parameters and variations in fluid and rock properties. Traditional methods require extensive computational resources to run numerous simulations for tasks such as history matching and probabilistic forecasting. The use of high-performance computing (HPC) simulators has been standard, but they demand significant computational power.

Innovative Approach with NVIDIA Modulus

Stone Ridge Technology, through its ECHELON reservoir simulator, has developed a scalable framework that integrates with NVIDIA Modulus on AWS. This integration leverages machine learning (ML) techniques, specifically neural operators, to create full-field proxy models. These models are significantly faster than traditional simulations, offering speed increases of 10x to 100x while maintaining accuracy.

The process involves ECHELON generating large datasets stored on Amazon S3, which are then used to train the ML models. The resulting proxy models can perform solution inference much faster than forward simulations, broadening the scope of potential applications, including uncertainty quantification and field optimization challenges.

Neural Operator-Based Proxy Generation

The use of a Fourier neural operator (FNO) model is central to generating spatio-temporal reservoir proxies. The FNO model, integrated with NVIDIA Modulus, allows for high-fidelity approximations of reservoir behaviors over time and space. This methodology provides a robust framework for addressing complex geological scenarios.

Implementation and Results

Implemented on AWS, the system utilizes flexible, on-demand computing resources to handle large data volumes. Stone Ridge's approach has shown promising results in various scenarios, including well placement optimization and geological uncertainty assessments. The models demonstrate a strong agreement with ground-truth simulations, showcasing minimal errors in pressure and saturation fields.

For example, in a well placement optimization scenario, the Modulus-based FNO proxy effectively captured the complex dynamics of water saturation in a reservoir, confirming its potential for real-world applications.

Conclusion

Stone Ridge Technology's integration of NVIDIA Modulus with its ECHELON simulator marks a significant advancement in reservoir simulation technology. By providing faster and more accurate simulations, this framework has the potential to transform subsurface applications, offering enhanced performance and reliability in energy exploration.

For more information on this development, visit the NVIDIA blog.