- Machine learning approach to predict the mineral compositions using conventional and geochemical well logs
- The goal of this study is using a combined approach with previous U-Net and XGBoost algorithm for mineralization composition analysis with petrophysical logs from selected formations. The code has been tested without any issue in PyCharm IDE.
- Install python 3.8/newer and anaconda packages: conda install keras-2.8.0, tensorflow-2.8.0, scipy, scikit-image, tqdm, pandas, numpy, seaborn, shutil, matplotlib, lasio etc. libraries.
- The original datasets of conventional and geochemical logs are from Horn River Basin. The loaded data are edited and used for testing purpose only.
-
The three main python codes and /itools, /data need be save in same folder. Before run the code, need install open-source Python packages.
-
forWeb1_Data_prepare.py is used to loading '.las' files, generate the training dataset, and save as '.csv/.pkl'.
-
forWeb2_Model_training.py is used to create the models, and training/validation with datasets from step 2. All of the weight parameters will be saved in one folder, code will run through each of the components and save all of the results in one folder.
-
forWeb3_Prediction.py is used for predicting mineral composition with conventional petrophysical logs, and plotting output figures.
For pilot running, user does not need change any parameters, just run the code files in steps 3,4. The code should run without any error if environment setting is correct. Please contact at following email address if any bugs popup.
- HP-7920 workstation: 56core CPU; 64G memory; one Nvidia Quadro P5000 GPU.
Unless otherwise noted, the source code of this project is covered under Crown Copyright, Government of Canada, and is distributed under the MIT license
The Canada wordmark and related graphics associated with this distribution are protected under trademark law and copyright law. No permission is granted to use them outside the parameters of the Government of Canada's corporate identity program. For more information, see Federal identity requirements.
- This Mineralogical characterization study used open source codes and library from github, google, and open-sourced geoscience packages lasio. Please cite the related references in your publications.
- Hu, K., Liu, X., Chen, Z. & Grasby, S.E. 2023, "Mineralogical Characterization From Geophysical Well Logs Using a Machine Learning Approach: Case Study for the Horn River Basin, Canada", Earth and space science (Hoboken, N.J.), vol. 10, no. 12, pp. n/a.
