Application of machine learning in the lithofacies study of unconventional oil and gas reservoirs

This paper is the result of geological survey engineering.

Objective Lithofacies is an inherent property of reservoirs, and high-quality lithofacies serves as a crucial foundation for the development of geological sweet spots in oil and gas reservoirs. Traditional lithofacies research methods struggle to meet the demands of highly heterogeneous and complex unconventional reservoirs, posing significant challenges for the efficient and effective identification of reservoir lithofacies.

Methods This study employs a combined approach of theoretical research and literature analysis to systematically review relevant research achievements both domestically and internationally. It provides an in-depth analysis of the current application status and development trends of machine learning in lithofacies research from multiple perspectives, comprehensively exploring the emerging interdisciplinary field of "machine learning + lithofacies."

Results Leveraging the powerful data analysis capabilities of machine learning algorithms to efficiently uncover nonlinear relationships within complex geological data for lithofacies identification has become an important method and a hotspot in the field of geological research; Machine learning is widely applied in lithofacies identification, encompassing supervised learning single models, ensemble models, neural networks, deep learning, and unsupervised models, providing diverse technical means for high-precision identification of reservoir lithofacies.; The practical application of machine learning methods in lithofacies identification faces challenges such as data quality, model interpretability, and computational resources. Future research should focus on advancing multi-source data fusion, model interpretability, and large-scale model technologies.

Conclusions Against the backdrop of China's oil and gas exploration advancing into deep and ultra-deep layers, exploration and development face significant challenges. By integrating machine learning technology to conduct efficient and intelligent research on underground lithofacies, we can accurately characterize the spatial distribution patterns of lithofacies and delineate lithofacies sweet spots. This will provide more precise and efficient technical support and decision-making guidance for the exploration and development of unconventional oil and gas resources.