Application of neural architecture search in lithology identification

Abstract Identifying rock types is the essential step in geological exploration because it guides reservoir description and development planning. Conventional methods that rely on empirical correlations or elementary machine learning approaches frequently produce suboptimal outcomes under complex, m...

Full description

Saved in:
Bibliographic Details
Main Authors: Yuhao Zhang, Hanmin Xiao, Meng Du, Qingjie Liu, Jingwei Tao, Yongcheng Luo, Li Peng, Jianbo Tan
Format: Article
Language:English
Published: SpringerOpen 2025-07-01
Series:Journal of Petroleum Exploration and Production Technology
Subjects:
Online Access:https://doi.org/10.1007/s13202-025-02039-y
Tags: Add Tag
No Tags, Be the first to tag this record!
Description
Summary:Abstract Identifying rock types is the essential step in geological exploration because it guides reservoir description and development planning. Conventional methods that rely on empirical correlations or elementary machine learning approaches frequently produce suboptimal outcomes under complex, multidimensional subsurface conditions. Although recent advancements in Artificial Intelligence (AI) have introduced automated approaches, these often exhibit limited adaptability when confronted with intricate well-log data. To address these constraints, the present study proposes an enhanced Neural Architecture Search (NAS) framework featuring an expanded search space that incorporates advanced the deep learning constructs, including one-dimensional deep neural network (DNN), long short-term memory (LSTM), and Transformers. This comprehensive strategy facilitates the automated discovery of specialized network configurations specifically designed for lithology classification. Experimental findings indicate that the NAS-derived model achieves an accuracy of approximately 96% on lithology test data, underscoring its effectiveness in managing heterogeneity within complex formations. Furthermore, the incorporation of Shapley Additive Explanations (SHAP) enhances interpretability by quantifying the contribution of each logging parameter, thereby ensuring consistency with geological reasoning. These results highlight the potential of a geoscience-adaptive NAS methodology for lithology identification by delivering improved performance, greater adaptability, and reduced reliance on exhaustive manual tuning.
ISSN:2190-0558
2190-0566