Application of machine learning models for real-time prediction of the formation lithology and tops from the drilling parameters

Lithology changes significantly affect the drilling program and the total cost of drilling an oil well, therefore, it is very important to detect the lithology variation and formation tops while drilling. Nowadays, the lithology changes are estimated from the geological data, which increases the uncertainty, especially during the exploration phase because of the data limitation. This study investigates the ability to use three machine learning models, namely, the artificial neural networks (ANN), adaptive neuro-fuzzy inference system (ANFIS), and functional neural networks (FNN) to predict the lithology changes and the formation tops in real-time while drilling. These models were trained on 3162 datasets of six inputs (drilling parameters) to enable real-time prediction of the lithology changes and formation tops through four different formations of sandstone at top, followed by anhydrite, then carbonate with shale streaks, and finally carbonate formation. The optimized models were then tested on 1356 datasets gathered from a well other than the training well. The results of this study confirmed the high accuracy of the optimized ANN model which outperformed both ANFIS and FNN models in predicting the actual lithology distribution and formation tops through both training and testing wells. The ANN model predicted the different four formations considered in this study with an accuracy of more than 98.1% for training and testing data. Although the ANFIS and FNN models predicted the carbonate/shale formation top accurately for the training data, they were not able to predict the top of all formations accurately for the testing data.