Abstract

In the context of the petroleum industry increasingly emphasizing the optimization of geological data analysis and interpretation to enhance exploration and production efficiency, machine learning techniques have demonstrated significant potential in processing and analyzing large volumes of complex, high-dimensional geophysical data. The advancement of unsupervised learning approaches has introduced new opportunities for extracting hidden knowledge from well log datasets without relying solely on subjective expert interpretation. Among these approaches, the Self-Organizing Map (SOM) algorithm has emerged as an effective tool due to its capability to perform clustering while preserving the nonlinear relationships among input variables and visualizing multidimensional data in an interpretable two-dimensional representation.

SOM has been widely applied in various petroleum engineering applications, including lithology classification, sedimentary environment identification, hydraulic flow unit delineation, reservoir property prediction, well log interpretation, and reservoir modeling. By projecting complex data structures onto a topological map, this algorithm enables geoscientists and reservoir engineers to identify characteristic patterns, reveal relationships among geophysical parameters, and support decision-making processes under conditions of data complexity and uncertainty.

This paper presents the application of the Self-Organizing Map algorithm for clustering well log data to identify potential lithological groups. The input dataset consists of selected conventional well logs, which were preprocessed and normalized prior to SOM training and analysis. The results demonstrate that SOM effectively separates data into clusters with similar characteristics, thereby capturing reservoir heterogeneity and facilitating the identification of intervals with potential hydrocarbon-bearing properties. The findings confirm the feasibility and effectiveness of SOM as a tool for geophysical data analysis and provide additional scientific support for well potential assessment and prediction, ultimately improving the reliability of decision-making during petroleum exploration and development activities.