Well Logging Based Lithology Identification Model Establishment Under Data Drift: A Transfer Learning Method.

Area of Science:

• Geoscience and Artificial Intelligence
• Machine Learning Applications in Petrophysics

Background:

• Well logging data often exhibits data drift due to variations in sedimentary environments and logging techniques.
• Existing machine learning models trained on historical well data perform poorly on new wells because of differing data distributions.
• This necessitates advanced methods to adapt models to new geological contexts.

Discussion:

• The proposed data drift joint adaptation extreme learning machine (DDJA-ELM) integrates project mean maximum mean discrepancy, joint distribution domain adaptation, and manifold regularization.
• This approach aims to bridge the distribution gap between source (old wells) and target (new wells) datasets.
• The method enhances the robustness and applicability of machine learning models in dynamic geological settings.

Key Insights:

• DDJA-ELM significantly improves the accuracy of lithology identification in new wells compared to traditional models.
• The technique effectively mitigates the negative impact of data drift on predictive performance.
• Experimental validation across multiple wells in the Jiyang Depression confirms the method's efficacy.

Outlook:

• This research offers a promising solution for enhancing the reliability of AI-driven lithology identification in the oil and gas industry.
• Future work could explore the application of DDJA-ELM in other geological prediction tasks.
• Further refinement of transfer learning techniques will be crucial for advancing autonomous geological interpretation.