The dataset contains realizations of geological stratigraphic curves which were generated using the included script.
This data is required to replicate results in Sergey Alyaev and Ahmed H. Elsheikh. "Direct multi-modal inversion of geophysical logs using deep learning." arXiv preprint arXiv:2201.01871 (2021).
This stratigraphy-realization dataset consists of randomly generated stratigraphic vertical depth functions b∗(x) which follow a known trend (here zero).
They need to be combined with an offset well-log, e.g. gamma-ray log from the Geosteering World Cup: https://doi.org/10.18710/20VIVT.
The training data consists of triples: a reference offset-well log which is trimmed
randomly to a short section of 64 cells (32 feet TVD); a sample of b∗(x) with 32 points
(32 feet); and an observed well-log corresponding to the first 16 feet of b∗(x), obtained
using the code supplied in trajectories_data_set.py.
The full training dataset, if read with overlap, contains 28 million samples, stored in train.nc. Additionally, we use a test dataset generated with the same rules containing 560 thousand samples, stored in test.nc.
Abstract of the publication
Geosteering of wells requires fast interpretation of geophysical logs which is a non-unique
inverse problem. Current work presents a proof-of-concept approach to multi-modal probabilistic inversion of logs using a single evaluation of an artificial deep neural network
(DNN). A mixture density DNN (MDN) is trained using the ”multiple-trajectory-prediction”
(MTP) loss functions, which avoids mode collapse typical for traditional MDNs, and allows multi-modal prediction ahead of data. The proposed approach is verified on the real-time stratigraphic inversion of gamma-ray logs. The multi-modal predictor outputs several likely inverse solutions/predictions, providing more accurate and realistic solutions
compared to a deterministic regression using a DNN. For these likely stratigraphic curves,
the model simultaneously predicts their probabilities, which are implicitly learned from
the training geological data. The stratigraphy predictions and their probabilities obtained
in milliseconds from the MDN can enable better real-time decisions under geological uncertainties.
pytorch, 1.10.2
python, 3.7
numpy, 1.21.2
matplotlib, 3.5.1
