Details:
Faculty Member: Eduardo Gildin
Department: Harold Vance Department of Petroleum Engineering
Abstract
The quest for fast simulation models has spurred innovation in computational methodologies, driving the development of reduced-order models leveraging machine learning architectures through data-driven simulations. Although successful implementation of non-intrusive techniques has emerged in reservoir simulation, it still poses challenges related to its generalizations and interpretability. In this talk, we propose a novel approach to Operator Inference (OpInf), which is a nonintrusive method to compute projection-based reduced-order model from snapshot (simulated or measured) data. I will show the benefits and some shortcomings in the application of reservoir simulation and will compare with AI-based solutions such as FNO’s and Deep Koopman.