Unsupervised Learning Based Lithology Classification from Well Log Data: A Primary Focus on Self-Organizing Map

Well logging is the process of obtaining information regarding the subsurface through boreholes. It measures properties such as density, porosity, fluid saturation etc, which are useful in identifying and classifying the various lithologies within the subsurface. Lithology classification and identification are crucial for reservoir characterization and oil and gas exploration. However, conventional methods, such as core sampling, are time consuming and expensive. In this research, a method of classifying lithology from well log data is developed using an unsupervised machine learning algorithm, self-organizing map (SOM). Various input features are considered to train the model, and lithology classification predictions are made and compared with pre-existing lithology data to evaluate the prediction accuracy. To minimize the impact of hyperparameters, we employ an ensemble approach by constructing the SOM 100 model. This proposed method aims to reduce the uncertainty associated with a single model and enhance the reliability of lithology classification prediction.