Evaluation of Machine Learning Techniques for Missing Well Log Data in Buzurgan Oil Field: A Case Study

The investigation of machine learning techniques for addressing missing well-log data has garnered considerable interest recently, especially as the oil and gas sector pursues novel approaches to improve data interpretation and reservoir characterization. Conversely, for wells that have been in operation for several years, conventional measurement techniques frequently encounter challenges related to availability, including the lack of well-log data, cost considerations, and precision issues. This study's objective is to enhance reservoir characterization by automating well-log creation using machine-learning techniques. Among the methods are multi-resolution graph-based clustering and the similarity threshold method. By using cutting-edge machine learning techniques, our methodology shows a notable improvement in the precision and effectiveness of well-log predictions. Standard well logs from a reference well were used to train machine learning models. Additionally, conventional wireline logs were used as input to estimate facies for unclassified wells lacking core data. R-squared analysis and goodness-of-fit tests provide a numerical assessment of model performance, strengthening the validation process. The multi-resolution graph-based clustering and similarity threshold approaches have demonstrated notable results, achieving an accuracy of nearly 98%. Applying these techniques to data from eighteen wells produced precise results, demonstrating the effectiveness of our approach in enhancing the reliability and quality of well-log production.