Deep Learning Applications for Well Log Correlation

This paper presents a novel approach to well-log correlation analysis, leveraging the capabilities of autoencoders within a deep learning framework. We explore the application of both single and dual-log data inputs to autoencoders, aiming to enhance the correlation analysis of geological layers in oil and gas exploration. Our methodology involved the utilization of Log ASCII Standard (LAS) files, containing various log types, such as Gamma Ray from Resistivity (GRR), Resistivity Deep (RD), Bulk Density (ZDEN), and Compensated Neutron Porosity Sandstone (CNC). The research was conducted across two datasets, each presenting different combinations of these logs. Key findings demonstrate that while single-log autoencoder models provide valuable insights, the integration of dual-log data offers a valuable alternative in scenarios where the most suitable log type for a particular geographical context is unknown. Furthermore, we present a comparative analysis of the correlation accuracy between different log types and their combinations. This analysis highlights the importance of log selection based on geological context and underscores the variability in performance among different logs. In summary, our research contributes to the field of geological analysis by proposing an autoencoder-based approach that effectively utilizes the power of multiple well-log data types. This methodology not only streamlines the well-log correlation process, but also provides geological insights for geoscientists. They can utilize both single-log and dual-log data inputs to enhance the accuracy, depth, and richness of geological interpretations in oil and gas exploration.