Graph-Based Methods for Reduced-Order Reservoir Simulation

Explore graph-based approaches for reduced-order reservoir simulation in this 59-minute lecture presented by Knut-Andreas Lie from SINTEF. Delve into the comparison between two specific methods: the interwell family, exemplified by GPSNet, and the CGNet approach. Discover how these models represent reservoirs, with GPSNet using 1D interwell connections and CGNet mimicking intercell connections of traditional 3D grids. Learn about the training process for matching well responses from fine-scale simulations and understand why CGNet may be easier to set up and generalize better than GPSNet. Gain valuable insights into making field management optimization and uncertainty quantification more computationally feasible through reduced-order models.