.png){kind=icon}
Return
A zero-training framework for facies classification using transformer-based vector embeddings
.png){kind=icon}
OA
AI
DOI：10.1016/j.aiig.2026.100216.png)
.png){kind=icon}
.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
Efficient subsurface drilling operations require rapid classification of changing lithology and facies for casing point selection, adjusting drilling fluid, and optimizing surface parameters. We present a zero-training framework that converts well log measurements into concise, domain-specific textual descriptions, then encodes them into high-dimensional vector representations using a transformer model (OpenAI text-embedding-3-large, 256 dimensions). Facies are then classified by maximizing the cosine similarity score between the test sample and reference embedding library, achieving classification accuracies up to 66% on blind wells and outperforming a tuned LightGBM benchmark, without gradient-based learning or transformer weight updates. The approach yields low-latency inference and a compact reference library, making it suitable for on-rig deployment and for straightforward adaptation to new fields. Beyond drilling, the approach can be generalized to any high-volume workflow that demands rapid and accurate classification in high-dimensional space.
Keywords:
Facies
Lithology
Transformers
Embeddings
Drilling
Classification
Machine learning
AI Summary
Key information extracted from the uploaded paper, including a brief overview, abstract, background, key highlights, visual analysis, and future outlook.
Journal
A
IF:
4.2
Papers:
46
Citations:
0
Organization
Cited Papers
ShareSaveShareSaveReal-Time Drilling Performance Optimization Using Automated Penetration Rate Algorithms with Vibration Control.png)
fuels
OAAI
fuelsIF0
OAAIShareSaveShareSaveShareSave no more
no more