Finding Low-Dimensional Representations of the Subsurface Earth Using Deep Learning
Large geological models are needed for modeling the subsurface processes in geothermal, carbon-storage, and hydrocarbon reservoirs. The size of these models contributes to the computational cost of history matching, engineering optimization, and forecasting. To reduce this cost, low-dimensional representations need to be extracted. Deep-learning tools, such as autoencoders, can find these geologically consistent low-dimensional representations.
Source: spe.org