Petrophysical properties prediction from prestack seismic data using convolutional neural networks

Abstract

We build Convolutional Neural Networks (CNNs) to obtain petrophysical properties in the depth domain from pre-stack seismic data in the time domain. We compare two workflows – i) end-to-end CNN (PetroNet) – to directly predict petrophysical properties from prestack seismic data, and ii) cascaded CNNs with two CNN architectures – the first network (ElasticNet) to predict elastic properties from pre-stack seismic data followed by a second network (ElasticPetroNet) to predict petrophysical properties from elastic properties. Cascaded CNNs with more than twice the number of trainable parameters as compared to end-to-end CNN showed similar prediction performance for a synthetic dataset. The average correlation coefficient for test data between true and predicted clay volume (around 0.7) is higher than the average correlation coefficient between true and predicted porosity (around 0.6) for both the networks. The cascaded workflow depends on the availability of elastic properties and is three times more computationally expensive than the end-to-end workflow for training. Coherence plots between the true and predicted values for both cases show that the maximum coherence occurs for values of inverse wavenumber above 15 m, which is approximately equal to 1/4 the source wavelength or λ/4. The network predictions have some coherence with the true values even at a resolution of 10 m, which is half of the variogram range used in simulating the spatial correlation of the petrophysical properties. The Monte Carlo dropout technique is used for approximate quantification of the uncertainty of the network predictions. An application of the end-to-end network for prediction of petrophysical properties is made with the Stybarrow field located offshore Western Australia. The network made good predictions of petrophysical properties at the well locations. The network is particularly successful in identifying the reservoir facies of interest with high porosity and low clay volume.

Deep Learning for Geophysics: Current and Future Trends

“…Therefore, DNNs for image classification can be directly applied in seismic attribute analysis (Das et al 2019, You et al 2020). If the attributes cannot be directly computed from the seismic data, a DNN can work in a cascaded way (Das and Mukerji 2020). If labels are not available, CAE is used for feature…”

“…Therefore, DNNs for image classification can be directly applied in seismic attribute analysis (Das et al, 2019;Feng, Mejer Hansen, et al, 2020;You et al, 2020). If the attributes cannot be directly computed from the seismic data, a DNN can work in a cascaded way (Das & Mukerji, 2020). If labels are not available, CAE is used for feature extraction, and then a clustering method, such as K-means, is used for unsupervised clustering (Duan et al, 2019;He et al, 2018;Qian et al, 2018).…”

Deep learning a poroelastic rock-physics model for pressure and saturation discrimination

“…Currently, there is a fast-growing application of deep neural networks to support interpretation and derive elastic properties from seismic data (Grana et al, 2017;Araya-Polo et al, 2018;Wu and Lin, 2018;Biswas et al, 2019;Das et al, 2019;Zheng et al, 2019;Das and Mukerji, 2020). Applications of machine learning have long been constrained by limiting computational capacities.…”

Seismic inversion with deep learning

“…The specific focus of our work is on the task of supervised classification, for which a brief, general introduction can be found in [22]. Convolutional neural networks (CNNs) have been successfully used as supervised classifiers for predicting physical properties of rocks such as seismic impedance [8], subsurface elastic parameters [35], or petrophysical properties [7]. In this work we also consider a set of deep-learning-based methods, that learn through a CNN the Fig.…”