Treffer: Application of Machine Learning for Prediction of Formation Lithology.

Evaluation and confirmation of formation lithology is vital for oil and gas properties analysis and hydrocarbon assessment. Prediction of formation lithology with a developed model will help to reduce cost of continuously logging within the same geological zone and terrain. In this work, supervised machine learning approach was utilized to predict formation lithology. Petrophysical properties and well log data from two wells (well Y and well Z) of North Sea were cleaned, splitted (trained and tested) and transformed. The data were inputted into python after coding and Microsoft excel to plot the data and established a diagnostic workflow. Random Forest Algorithm (Baseline model) was used as a meta estimator by fitting multiple randomized trees on different subsets of the data set and averaging. A supervised machine learning model was developed, trained, evaluated and tested with mean absolute error and root mean square error. Well log lithology for shale and sandstone were examined with the prediction of the developed model. Results show that the developed model had a better prediction for sandstone formation when compared with the actual. Similarly, the predicted values yielded less errors than the actual values. Also, robust performance of supervised learning models in predicting lithologies of both wells reflects their effectiveness in handling the complexity of fluid properties in the reservoir. The built model proved accurate to predict lithology of sandstone, however a hybrid model should be tested, and rigorous cross-validation techniques be implemented to ensure that models generalize well to unseen data and are less prone to overfitting. [ABSTRACT FROM AUTHOR]

Copyright of Improved Oil & Gas Recovery is the property of Smart Science & Technology and its content may not be copied or emailed to multiple sites without the copyright holder's express written permission. Additionally, content may not be used with any artificial intelligence tools or machine learning technologies. However, users may print, download, or email articles for individual use. This abstract may be abridged. No warranty is given about the accuracy of the copy. Users should refer to the original published version of the material for the full abstract. (Copyright applies to all Abstracts.)