There are many challenges associated with achieving carbon storage at gigaton scales. In this talk, I will present some of our recent developments in two areas relevant for the reservoir engineering of CCUS projects. A general framework for optimizing CO2 storage operations using derivative-free algorithms will be described. Different objective functions (involving minimization of mobile CO2 and maximization of storage efficiency) will be considered, along with a range of practical constraints. A multifidelity optimization treatment will be shown to be effective and to provide improved computational efficiency. The use of a deep-learning-based surrogate model for history matching will then be discussed. A previously developed architecture, referred to as a recurrent-residual U-Net, is extended to treat coupled flow-geomechanics problems. Its ability to model pressure and plume location at the Earth’s surface will be demonstrated. Finally, the deep-learning surrogate will be applied within a history matching workflow.
Professor Louis J. Durlofsky is the Otto N. Miller Professor of Earth Sciences in the Department of Energy Science and Engineering at Stanford University. He codirects the Stanford Smart Fields Consortium and the Stanford Center for Carbon Storage. Earlier in his career, Durlofsky was affiliated with Chevron Energy Technology Company. He holds a BS degree from Pennsylvania State University, and MS and PhD degrees from the Massachusetts Institute of Technology, all in chemical engineering. His research interests include subsurface flow simulation and optimization, history matching, uncertainty quantification, deep-learning-based surrogate modeling, and energy systems optimization.