Real-Time Lithology Prediction While Drilling Using Machine Learning Algorithms: A Web Application Based Solution

Abstract

Abstract Formation lithology is a vital information source for initial well economic assessment, reservoir lithology description, geological correlation, formation identification, verification of wireline log response, and identification of hydrocarbon deposits. Real-time lithology determination while drilling can greatly optimise the drilling process and enhance geosteering. Using linear, non-linear, and ensemble machine learning algorithms, this study sought to automate the process of lithology classification while drilling. Also, the project sought to create a web application with a graphical user interface to enable the users easily train and save models which can be applied on different lithology classification projects without the need to code. A public dataset containing mudlog and lithology data from ten wells in the Volve oil field, Norway, was used in this study. The graphical user interface was then created using Streamlit and Python programming language to allow for easier application of the machine learning algorithms without a prior programming knowledge. The results showed that the ensemble methods (Random Forest classifier and Gradient Boosting classifier) outperformed the other linear (logistic regression) and non-linear (Support Vector) machine learning algorithms. Random Forest classifier achieved an overall accuracy of 99% and a precision of 98% while the Gradient Boosting classifier had an accuracy score of about 97% and a precision of 92% in classifying the six lithology types from ten wells. The results obtained showed that implementing ensemble machine learning algorithms with a graphical user interface can assist Engineers to classify lithology efficiently, easily, and timely while drilling. This web application can also be used by geologists and geophysicists to classify lithology using machine learning without the need to write codes.