Session: 11-03-03 Data Science Applications in Drilling

Paper Number: 123362

123362 - Unsupervised Machine Learning for Well Log Depth Alignment 

Well-log data is essential for understanding subsurface geological structures. Logging companies collect substantial volumes of well-log data from each well. However, this raw information requires processing and alignment to extract meaningful insights effectively. Until now, manual well-log depth alignment based on correlation has been a popular yet tedious and time-consuming process in the field. This research explores the application of unsupervised machine learning techniques to the problem of well-log depth matching, providing an innovative solution to the challenges in aligning data from Logging While Drilling (LWD) and Electrical Wireline Logging (EWL) from the same well.

The primary objective of this study is to align LWD logs with the reference EWL logs from the same well, ensuring they share a common depth reference. We utilize key well-log measurements, such as the Gamma Ray (GR), resistivity, density, and neutron porosity logs to achieve this. These foundational well-log attributes hold a crucial role in predicting the petrophysical properties of the well and in enhancing our understanding of lithological variations within the subsurface.

The methodology employed in this research involves the utilization of simple yet effective unsupervised machine learning libraries, such as K-means clustering, Density-based clustering (DBSCN), and fuzzy C-means clustering together with Principal Component Analysis (PCA), and autoencoders.

Our study demonstrates promising results by employing a combination of unsupervised machine-learning techniques to align LWD and EWL well logs. Initially, we utilize clustering algorithms, such as K-means, Density-based clustering (DBSCN), and fuzzy C-means, to create clusters representing groups of similar data points from both EWL and LWD data. While these clusters do not directly correspond to lithology groups, they serve as a foundational step for depth alignment, marking a significant advancement in unsupervised well-log depth alignment.

To further enhance the alignment process, we apply dimensionality reduction methods, such as Principal Component Analysis (PCA), which effectively reduces the dimensionality of the data. After dimension reduction, we continue to use clustering techniques, complementing the initial clusters with the refined data representation. Additionally, we employ autoencoders, a neural network architecture, to extract compressed data from the bottleneck layer, which is subsequently subjected to clustering techniques.

This multifaceted approach significantly improves our understanding of subsurface geology by aligning the group boundaries between LWD and EWL. Moreover, it has immediate practical implications for subsurface exploration and geological studies, enabling more accurate lithology predictions. Our contribution presents a novel application of unsupervised machine learning for well-log depth alignment, contributing to ongoing developments in geophysical data analysis in the oil and gas industry. These innovative methods offer valuable potential benefits in reservoir characterization and industry decision-making processes.

While specific results are not detailed in this abstract, our study has shown promising outcomes in accurately aligning well logs using unsupervised machine-learning techniques. The comparative analysis suggests that these techniques have the potential to improve depth-matching precision, enhancing our understanding of subsurface geology.

Presenting Author: Sushil Acharya sushil.acharya@ntnu.no

Presenting Author Biography: Sushil Acharya, originally from Nepal and now residing in Norway, holds two master's degrees. His first master's in Mathematics and Physics from Tribhuvan University in Nepal laid the foundation for his academic journey. Pursuing a second master's at Oslomet University, Norway, with a specialization in Data Science, Sushil has honed his skills in extracting valuable insights from data.

Currently a Ph.D. candidate at the Norwegian University of Science and Technology (NTNU), Sushil's research centers on dynamic wellbore measurements alignment using machine learning, a vital aspect in the realm of oil and gas exploration.

With a diverse background and a passion for data science, Sushil Acharya is poised to make significant contributions to the academic and research community.

Authors:

Sushil Acharya sushil.acharya@ntnu.no
Karl Fabian Norwegian University of Science and Technology