Machine learning has been an emerging technique in dealing with various geoscience problems. One of the most popular applications is to conduct electrofacies classification using both unsupervised (clustering) and supervised learning methods which have been discussed in great details in the set of below reference papers. Some papers have used solely supervised machine learning with training data obtained from actual data such as core or FMI while others demonstrate the use of a set of machine learning algorithms from dimensional reduction (PCA) to clustering (unsupervised) with calibration to measured data and finally using supervised learning to generate electrofacies or to crosscheck the results from the previous step. This is a more advanced workflow and has appeared to perform very well in highly heterogeneous clastics and carbonate reservoirs from around the world. The use of clustering technique with calibration of measured data allows geoscientists to conduct electrofacies classification with limited training dataset which is a common issue in petroleum geoscience. 

Thorough discussions on the machine learning algorithms in these papers provide great insights into mathematical aspect of the technique. For supervised learning, Random Forest Classification which is a tree-based learning approach stands out to be of one the most efficient machine learning methods for electrofacies classification mainly because merging results from several trees will avoid overfitting and can better generalize to new data. K-mean clustering, on the other hand, appear to be a powerful unsupervised machine learning technique for electrofacies classification using wellbore data. Together with Principal Component Analysis, they should be added to any conventional data analysis toolkit for wellbore data. Apart from the algorithms, it would be great to see more discussions on how to choose input data and model configurations specifically for each data set because they are critical steps in building a successful machine learning model. Without doing model validation or blind test, we are at risk of being underfitting or overfitting. Therefore, in case the available data do not allow us to have a meaningful validation there should be at least a geological interpretation of the prediction results which are almost missing in these papers. Incorporation of geological components into the workflow increases reliability of predicted results. Another way to do this is to use extracted features from original logs such GR patterns or to add simple rule-based condition to refine the machine learning results. It will be much more valuable if the authors go further by making electrofacies prediction and put all wells into a geological framework to see how machine learning can assist in building a reservoir model. For example, how classification results improve permeability prediction and ultimately production history matching and optimization.

The machine learning algorithms were currently performed on sample-by-sample basis which might not be the best approach in dealing with geological problem mainly because the succession consists of many rock units that are related to each other by, for example, being deposited in the same depositional system. Therefore, it is recommended that geoscientists should try a research direction of defining rock units first and then build a machine learning model to make prediction on unit-by-unit basis.

References:

  1. Paolo Bestagini, Vincenzo Lipari and Stefano Tubaro, 2017, “A Machine Learning Approach to Facies Classification Using Well Logs”, SEG International Exposition and 87th Annual Meeting.
  2. Watheq J Al-Mudhafar, 2020, “Advanced Supervised Machine Learning Algorithms for Efficient Electrofacies Classification of a Carbonate Reservoir in a Giant Southern Iraqi Oil Field”, Offshore Technology Conference OTC-30906-MS
  3. Watheq J Al-Mudhafar and Erfan M Al Lawi, 2019, “Clustering Analysis for Improved Characterization of Carbonate Reservoirs in a Southern Iraqi Oil Field”, Offshore Technology Conference OTC-29269-MS
  4. Wenhao Zheng, Fei Tian, Qingyun Di, Wei Xin, Fuqi Cheng and Xiaocai Shan, 2021, “Electrofacies classification of deeply buried carbonate strata using machine learning methods: A case study on ordovician paleokarst reservoirs in Tarim Basin”, Marine and Petroleum Geology 123 (2021) 104720
  5. Alexsandro G. C., Carlos A. C. da P., & Geraldo G. N, 2017, “Facies classification in well logs of the Namorado oilfield using Support Vector Machine algorithm”, 15th International Congress of the Brazilian Geophysical Society, Brazil, August 2017
  6. Touhid Mohammad Hossain, Junzo Watada, Izzatdin A. Aziz, and Maman Hermana, 2020, “Machine Learning in Electrofacies Classification and Subsurface Lithology Interpretation: A Rough Set Theory Approach”, Applied Science. 2020, 10, 5940

RELATED ARTICLESMORE FROM AUTHOR