Development of CRM for Reservoir Simulations Using PINNs | MathWorks Energy Conference 2022

Reservoir simulators play a key role in the management and optimal production from oil and gas fields. However, the computational costs of detailed simulations can be prohibitively expensive and most certainly not useful for real-time decision-making. In this presentation, a reduced-order model (ROM) is built using the time-series production data from a real oil and gas field. The CRM is chosen here as a reduced-order representation for the reservoir simulator.

With the increase in computational power and recent machine learning (ML) approaches, it is apparent that the oil and gas industry will eventually adopt useful models through proper validation. PINNs are the neural networks that can enforce the governing equations for the underlying dynamics as a part of building ML models. Results are compared against a detailed reservoir simulation to demonstrate the usefulness of ML models.