/CNN_based_petrophysical_inversion

Neural network based petrophysical property inversion

Primary LanguageJupyter Notebook

Petrophysical properties prediction using CNNs

Neural network based petrophysical property inversion.

Paper reference:

Das, V. and T. Mukerji, 2019, Petrophysical properties prediction from pre-stack seismic data using convolutional neural networks: SEG Technical Program Expanded Abstracts 2019, 2328-2332.

(https://library.seg.org/doi/abs/10.1190/segam2019-3215122.1)

The main folders with final neural network architectures for the synthetic cases are:

  1. Petrophysical from elastic - This contains the cascaded neural network architectures (Final architectures used in paper are in the following jupyter notebooks in the Base_case folder)

    i. Petrophysical_properties_from_seismic-input-elasticnet-near-far-Ip,Vp_Vs_ratio-comparable_network.ipynb - This is the network with number of trainable parameters equivalent to the end-to-end network

    ii. Petrophysical_properties_from_seismic-input-elasticnet-near-far-Ip,Vp_Vs_ratio.ipynb - This is the network that gives the best results for the cascaded network case

  2. Petrophysical from seismic - This contains the end-to-end neural network architecture (Final architectures used in paper are in the following jupyter notebooks)

    i. Petrophysical_properties_from_seismic_near_far.ipynb - This is the final network for the end-to-end approach

    ii. Petrophysical_properties_from_seismic_near_far-Copy1.ipynb - This is the final network for Stybarrow field example

Data for the synthetic example is in Data_generation_base_case folder

Data for the Sybarrow field example is in Data_Stybarrow_field folder