Improving Geological Interpretation and Prediction for Hydrocarbon Exploration using ML-Based Approaches

Abstract

This research investigates the application of machine learning (ML) techniques to geological interpretation and hydrocarbon exploration, presenting three novel methodologies that progress sequentially: lithofacies classification, well log data imputation, and seismic noise reduction. The first approach presents an innovative self-supervised strategy for seismic noise reduction using Deep Convolutional Denoising (DCD) networks. This strategy circumvents the limitations of conventional supervised learning methods, predicting noise-free values using surrounding noisy samples. When applied to the real-world Kerry Seismic dataset from a New Zealand field, the DCD successfully reduces contamination from SHAP-based noise, random noise, and ground roll noise, while preserving signal integrity. The second study introduces a cutting-edge well log data imputation method using time-series deep learning models in Indonesia's West Natuna Basin. Long short-term memory (LSTM) gated recurrent unit (GRU), and bidirectional LSTM (Bi-LSTM) models are trained to impute missing Vp log data, with the LSTM model exhibiting the most promising results—a mean absolute percentage error (MAPE) of approximately 2.2% and an R-squared (R2) value of 94%. These findings suggest that deep sequence models hold potential for effective well log data imputation, enhancing decision-making within the oil and gas industry. The final study employs a supervised learning method for lithofacies classification within well-logging data, with an emphasis on coal facies prediction in Indonesia's Tarakan Basin. A comparison of various ML algorithms reveals that Random Forest and Gradient Boosting models excel, achieving accuracies of 87.49% and 87.01% respectively. However, difficulties arise when classifying coal facies, necessitating rock physics analysis to understand potential misclassifications. Despite these challenges, the ML approach demonstrates a high precision in lithofacies prediction, particularly within the Tarakan Basin. Collectively, these studies underscore the significant potential of ML in improving geological interpretation and hydrocarbon exploration, contributing valuable advancements to complex geoscience problems and informing decision-making within the oil and gas industry.