Rock-physics based Augmented Machine Learning for Reservoir Characterization

Abstract

Summary The challenge in adopting neural networks in the geosciences is the relative scarcity of labeled training data. This presentation demonstrates an approach to augment the amount of data used to train the neural network. Rock Physics theory is used to model the elastic parameter response due to changes in the rock and fluid properties of the local well control to generate a large number of pseudo wells. These pseudo wells are then used to model synthetic seismic gathers which are then used to train a Deep Neural Network (DNN). The trained DNN is then applied to the real dataset. Application of this workflow is shown for seismic reservoir characterization on a field in the North Sea producing commercial volumes of oil. The results are shown to have good continuity, are high in resolution which is compared to the prestack inversion approach.