A MACHINE LEARNING APPROACH TO THE PREDICTION OF LITHOLOGY FOR DRILLING OPTIMIZATION

ABSTRACT

Predicting the formation type and lithology in a well before drilling, or at least while drilling, is crucial to reducing drilling issues, including loss of circulation and kick, as well as to increasing drilling pace, bit optimization, and shale swelling prevention. The lithology can be ascertained by many means, such as log interpretation; however, there is no precise way to ascertain the lithology prior to or during drilling. The identification of formation type and lithology is a very complex subject for which there is yet no analytical solution. Machine learning/Artificial intelligence looks like it may be quite beneficial in this case. The establishment of intricate non-linear mappings between inputs and outputs is possible with machine learning. The conventional technique of lithology identification or prediction is expensive, costly, and time-consuming. Therefore, this thesis suggests an alternate method of lithology prediction based on the random forest algorithm and artificial neural network. This study utilized well log data to forecast the lithology and formation type of a given formation with a reasonable degree of accuracy. For the development of these models, 405809 data sets from two Niger Delta wells were integrated into artificial neural networks, random forest algorithm and subjected to quality control and data mining. The results indicate that the random forest method and neural networks can identify the kind of formation and the lithology with around 95% and 85% accuracy, respectively.