Application of a Convolutional neural network (CNN) is presented for an unconventional reservoir to simultaneously predict elastic, mineral volumes, geomechanical and reservoir properties from seismic and well data.
In the Barnett Shale, success requires identifying brittle, frackable, productive layers with high quartz content. Identification of brittle productive layers also requires reliable elastic properties volumes. Seven wells are used to generate hundreds of synthetic wells and associated seismic gathers. The resulting synthetic data are used to train the CNN algorithm. Density, P-wave and S-wave impedance are estimated, from which Young’s Modulus and Brittleness Index are calculated. The second CNN simultaneously estimates the effective porosity, clay, kerogen, and quartz volumes. A Bayesian lithofacies method was used to perform facies classification and generate facies probabilities.
The CNN results better resolve the zone of interest, show better lateral continuity and better match with the well control, including two blind wells, than estimates based on a more traditional machine learning workflow using multilinear regression and hand selected attributes.
The objectives of this study is to accurately predict unconventional reservoir properties from seismic and well data using convolutional neural networks (CNNs). There is a great interest in the use of Deep Learning (Goodfellow et al.,2016) to automate complex workflows and reduce project times. Downton and Hampson (2021) show that CNNs can more efficiently predict reservoir properties as compared to traditional machine learning methods (Hampson et al., 2001) which use seismic attributes as input. However, there is often insufficient training data (i.e. well control) to properly train a CNN. Downton et al. (2020) overcome this issue using a hybrid theory-guided data science (TGDS) model (Karpatne et al., 2017). A two-component model is built where the outputs of the theory-based component are used as inputs to the data science component. Rock physics theory is used to perturb the original well control to generate many synthetic wells spanning the range of the expected geology. Seismic theory is then used to model synthetic seismic gathers to create a Synthetic Seismic Catalogue that contains both the target logs and input features. The Synthetic Seismic Catalogue is used to train and validate the CNN. The trained CNN is then applied to the real dataset. Downton et al. (2020) demonstrate this workflow on a North Sea Oil field utilizing an unconsolidated (soft) sandstone rock physics model (RPM) (Dvorkin and Nur, 1996). This current work replaces the RPM with an inclusion model (Key and Xu, 2002) more suitable for unconventional reservoirs.
