Abstract
The results presented in this research demonstrate how the K-means clustering algorithm was
integrated with well logs to build a machine learning model for delineating hydrocarbon-bearing
zones of conventional reservoirs and estimating reservoir parameters from well logs. The well
log data used for training the model were loaded and necessary cleanup processes were
performed which entail the handling of missing values and the removal of outliers, and some
exploratory data analyses (EDA) were carried out including the plotting of numerical
distributions, followed by feature selection and scaling. All the aforementioned steps were taken
in order to transform the well logs to a format suitable for the K-means algorithm. After the data
transformation phase, the K-means algorithm was applied on the well logs to develop a machine
learning clustering model that grouped depth points together based on the similarity of their log
values. The clusters generated from the clustering process were evaluated to determine the one
that consistently and accurately predict the hydrocarbon reservoirs in the wells. Some equations,
such as Archie’s water saturation equation, were incorporated so that reservoir parameters
could be estimated by the program. The model, when tested on three wells, delineated about
twelve zones as potential hydrocarbon-bearing zones. These zones were manually interpreted to
assess the accuracy of the model and it was found that its precision in delineating the
hydrocarbon reservoirs was high. The significance of the research study is that the model
developed can be used to automate the task of delineating hydrocarbon reservoirs from well
logs.