Automatic Seismic to Well Tie

Author: Valentin Tschannen - Fraunhofer ITWM, Germany.

This package aims to provide basic utilities to automate the seismic to well tie process. Part of the package is dedicated to the creation and training of a neural network to perform wavelet extraction. It also contains utilities to create synthetic training data. The other modules are dedicated to the remaining operations (log processing, depth to time conversion), and use random search and bayesian optimization for the automatic tuning of hyper-parameters.

Package content

• README.md, Makefile, setup.py, environment.yml, main.sh
• scripts : shell scripts for the installation
• tests : unit/intergration tests
• documentation : documentation built during installation
• notebooks : jupyter notebooks demonstrating the use of the library
• experiments : config files to train a neural network and folder where results are saved
• wtie : python package

Installation

Linux

(TODO: docker) Python dependencies are listed in the file enivronment_linux.yml. Assuming that you installed the Anaconda platform and that your system's os is unix-like with the tool make, you can install the wtie package by running the following command in a shell console in your base environment (alternatively, take a look at the Makefile to see the steps that you need to reproduce to install on your system):

make install

This will create a conda environment named wtie and install the package (this may take several minutes, especially if you are going through a network adapter). If there were no problems, you are ready to work with the package!

Don't forget to activate the environment before working with the package:

conda activate wtie

Windows

(TODO: better Windows support) Python dependencies are listed in the file enivronment_windows.yml. For this guide I assume that you installed the Anaconda platform. Go to the main well-tie directory and go through the follwoing steps:

First, open an Anaconda prompt and create a new conda environement:

conda env create -f environment_windows.yml

This will create a conda environment named wtie.

Second, open a Windows command prompt and activate the environment:

conda activate wtie

Then, in the same prompt, install the python wtie package:

python setup.py develop

Unfortunatly, Windows does not seem to be able to streamline the installation of the noise package that we use to generate correlated synthetic reflectivity series. To install this package you should follow instructions given here.

Finally, from the same prompt, move to the tests folder and run the test suite:

pytest -v --basetemp="./tmp"

Don't forget to activate the environment everytime you want to work with the package:

conda activate wtie

Tutorial

A series of notebooks demonstrates the use of the library to perform an automatic well tie. The weights of a pretrained network as well as data coming from the open Volve and Poseidon datasets are provided in the data folder. NOTE that both datasets come with their own license.

Explanations about how to train your own neural network are given here.

Publications

A publication titled "Partial automation of the seismic to well tie with deep learning and Bayesian optimization" is available in open access in the journal Computers and Geosciences. It can be accessed via https://doi.org/10.1016/j.cageo.2022.105120 . You are free to download and modify the source code for your work. If you find this work useful, don't forget to cite our publication.