Abstract
The proposed paper introduces a novel machine learning-based approach that incorporates advanced data pre- and post-processing techniques to significantly enhance automated lithology classification using well log data.
Through rigorous testing of various machine learning algorithms and data science methods, the final approach emerged as the optimal solution. It integrates advanced well log data visualization, exploratory data analysis (EDA) for feature selection and engineering, the use of the Random Forest algorithm for data pre-processing and classification, and advanced post-processing techniques to refine the predictions of the machine learning model. These post-processing steps are particularly novel, as they address the unique challenges of lithology classification in the petroleum industry.
The approach was developed during a company-wide hackathon with over fifty participants from around the globe. The objective was to design an accurate, automated system for predicting lithology using real-world well log data. The dataset included measurements from twenty wells with twelve interpreted facies and ten additional wells for evaluation. The complexity of the data, such as missing logs and incomplete measurements, mirrored real-world challenges. The proposed approach excelled in this context, achieving the highest competition score and securing first place in the hackathon.
This paper presents a proven method for automated facies classification using machine learning, validated by real field data. It underscores the importance of integrating domain expertise with advanced data science techniques, particularly post-processing, to deliver value in the age of AI to the petroleum industry.
