2025 Energy HPC Conference

Advancing Reservoir Engineering through High-Performance Computing and Neural Operators on the Cloud

Abstract: Contemporary reservoir engineering applications demand extensive high-fidelity simulations that remain computationally intensive despite advances in high-performance computing. This work presents an integration of scientific machine learning with physics-based reservoir simulation through a scalable, cloud-based workflow utilizing Fourier Neural Operators (FNOs) and GPU-accelerated simulators. FNOs learn mappings between function spaces rather than Euclidean spaces, enabling superior generalization capabilities. The framework is validated using two synthetic 2-phase oil-water systems: a homogeneous case and a heterogeneous case with multi-scale property variations. Results demonstrate that our HPC-enabled FNO implementation achieves approximately 1000x speedup compared to traditional approaches while maintaining acceptable accuracy. Future work will address scaling challenges and enhanced applicability in production environments.

Authors: Karthik Mukundakrishnan (Stone Ridge Technology), Vidyasagar Ananthan (Amazon Web
Services), Dan Kahn (Amazon Web Services) and Dmitriy Tishechkin (Amazon Web Services)