Related Experiment Video
Updated: Nov 17, 2025
10:05
Integrated Field Lysimetry and Porewater Sampling for Evaluation of Chemical Mobility in Soils and Established Vegetation
Published on: July 4, 2014
14.5K
Active Domain Adaptation With Application to Intelligent Logging Lithology Identification.
IEEE Transactions on Cybernetics
|February 18, 2021
Summary
This study introduces a new framework for intelligent logging lithology identification, combining active learning and domain adaptation to improve model accuracy in new exploration wells despite data distribution differences.
Area of Science:
- Geoscience and Artificial Intelligence
- Machine Learning Applications in Reservoir Exploration
Background:
- Lithology identification is crucial for formation characterization and reservoir exploration.
- Intelligent logging lithology identification uses machine learning on well-logging curves but faces challenges with data distribution discrepancies between training and new wells.
Purpose of the Study:
- To develop a robust lithology identification model for target wells using limited target-labeled data and abundant source-labeled data.
- To address challenges of distribution misalignment, data divergence, and cost limitations in intelligent logging lithology identification.
Main Methods:
- Proposed a novel active adaptation for logging lithology identification (AALLI) framework integrating active learning (AL) and domain adaptation (DA).
- Designed a discrepancy-based AL and pseudolabeling (PL) module with an instance importance weighting module to manage uncertain target data and confident source data.
- Developed a reliability detecting module to enhance the accuracy of target pseudolabels.
Main Results:
- The AALLI framework effectively handles domain discrepancy and data divergence in intelligent logging lithology identification.
- The proposed method demonstrates superior performance compared to baseline methods in extensive experiments on real-world well-logging datasets.
- Successfully addressed challenges including cost limitations, distribution misalignment, and data divergence.
Conclusions:
- The AALLI framework offers a significant advancement in intelligent logging lithology identification, particularly for new exploration wells.
- The integration of AL and DA provides a powerful solution for leveraging limited labeled data in geoscientific applications.
- The study highlights the potential of advanced machine learning techniques to overcome data challenges in reservoir exploration.
