Comparative Analysis of the K-Nearest-Neighbour Method and K-Means Cluster Analysis for Lithological Interpretation of Well Logs of the Shushufindi Oilfield, Ecuador

Abstract

The lithological interpretation of well logs is a fundamental task in Earth science that can be accomplished with the application of various machine learning algorithms. The current investigation attempts to evaluate the performance of the K-nearest-neighbour Density Estimate (KNN) and K-means cluster analysis methods for predicting lithology in a dataset of logs measured in the siliciclastic reservoir of the Shushufindi Oilfield of Ecuador. The comparison of lithological interpretation is assembled using classical methods, such as qualitative interpretation and density-neutron cross plot. The lithological interpretation results showed that the supervised method KNN has a higher fitting level with the comparison interpretation data (87.3%, 1145 m predicted of 1311.1 m interpreted) than the results of the K-means method (71.6%, 939.7 m predicted of 1311.1 m interpreted). The geological nature of the reservoir creates a level of a discrepancy because of the near geophysical responses between limestone and intermedia grain size rocks. The possibility of controlling this in the KNN algorithm makes it preferable for usage in these types of reservoir lithological interpretation.