Automatic Reservoir Interpretation from Conventional Well Logs Using Stacking Machine Learning Technique

Log interpretation is usually a significant job for asset development with relative high uncertainties and low efficiency. Incorrect results are associated with several negative effects. Remarkable efforts have been made to develop machine learning method to solve this problem. However, with only a single algorithm, the performance may be limited. To maximize prediction accuracy, we present another option using stacking machine learning technique, within which different types of base machine learning algorithms function together through a mega - classifier. Compared with the best performed single algorithm, stacking algorithm performs better, achieving 85.3% in prediction accuracy and 81.2% in F1 score. These results suggest that this strategy could lead to better prediction and improve the prediction accuracy.