Self‐Supervised Pre‐Training and Few‐Shot Finetuning for Gas‐Bearing Prediction

Abstract Natural gas remains the only fossil energy with sustained production growth amid global carbon reduction efforts, with seismic methods crucial for its exploration and development. Traditional seismic inversion methods, which detect gas indirectly by modeling and inverting gas‐sensitive para...

Full description

Saved in:
Bibliographic Details
Main Authors: Long Han, Xinming Wu, Renjie Chen, Yunhua Shi, Zhanxuan Hu, Huijing Fang
Format: Article
Language:English
Published: Wiley 2025-06-01
Series:Journal of Geophysical Research: Machine Learning and Computation
Subjects:
Online Access:https://doi.org/10.1029/2025JH000631
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1849432371855622144
author Long Han
Xinming Wu
Renjie Chen
Yunhua Shi
Zhanxuan Hu
Huijing Fang
author_facet Long Han
Xinming Wu
Renjie Chen
Yunhua Shi
Zhanxuan Hu
Huijing Fang
author_sort Long Han
collection DOAJ
description Abstract Natural gas remains the only fossil energy with sustained production growth amid global carbon reduction efforts, with seismic methods crucial for its exploration and development. Traditional seismic inversion methods, which detect gas indirectly by modeling and inverting gas‐sensitive parameters, can introduce cumulative errors, whereas conventional deep learning methods struggle with few‐shot generalization due to limited labeled data. This paper presents a deep learning workflow for directly predicting gas saturation from multiple seismic attribute data, involving pre‐training on large‐scale unlabeled data and finetuning with few labeled data to address the few‐shot challenge. Our network includes a Depthwise CNN1D for multi‐attribute feature extraction, an iTransformer for feature fusion, and a predictor for outputting the target. We use the iTransformer's self‐attention mechanism to calculate attribute weights for selection, and through extensive experiments, we developed a windowed multi‐attribute data input method that incorporates neighboring information to ensure lateral consistency. Based on the geological understanding that different attributes computed from the same sample convey correlated features reflecting geological properties, we pre‐train the network on large‐scale unlabeled data using a self‐supervised learning strategy of attribute masking and recovery. This approach encourages the network to learn correlation features, thereby improving its generalization by familiarizing it with the data distribution across the entire study field. We then employ the LoRA finetuning method to adapt the pre‐trained model to gas saturation prediction with meager labeled data while preserving pre‐trained knowledge. We applied this method to field data in the South China Sea, achieving accurate and generalized gas saturation predictions.
format Article
id doaj-art-caf64762315a4ee18226d7ab2ad7aa00
institution Kabale University
issn 2993-5210
language English
publishDate 2025-06-01
publisher Wiley
record_format Article
series Journal of Geophysical Research: Machine Learning and Computation
spelling doaj-art-caf64762315a4ee18226d7ab2ad7aa002025-08-20T03:27:22ZengWileyJournal of Geophysical Research: Machine Learning and Computation2993-52102025-06-0122n/an/a10.1029/2025JH000631Self‐Supervised Pre‐Training and Few‐Shot Finetuning for Gas‐Bearing PredictionLong Han0Xinming Wu1Renjie Chen2Yunhua Shi3Zhanxuan Hu4Huijing Fang5School of Earth and Space Sciences University of Science and Technology of China Hefei ChinaSchool of Earth and Space Sciences University of Science and Technology of China Hefei ChinaShenzhen Branch of CNOOC China Limited Shenzhen ChinaShenzhen Branch of CNOOC China Limited Shenzhen ChinaSchool of Information Science and Technology Yunnan Normal University Kunming ChinaExploration Research Institute Anhui Provincial Bureau of Coal Geology Hefei ChinaAbstract Natural gas remains the only fossil energy with sustained production growth amid global carbon reduction efforts, with seismic methods crucial for its exploration and development. Traditional seismic inversion methods, which detect gas indirectly by modeling and inverting gas‐sensitive parameters, can introduce cumulative errors, whereas conventional deep learning methods struggle with few‐shot generalization due to limited labeled data. This paper presents a deep learning workflow for directly predicting gas saturation from multiple seismic attribute data, involving pre‐training on large‐scale unlabeled data and finetuning with few labeled data to address the few‐shot challenge. Our network includes a Depthwise CNN1D for multi‐attribute feature extraction, an iTransformer for feature fusion, and a predictor for outputting the target. We use the iTransformer's self‐attention mechanism to calculate attribute weights for selection, and through extensive experiments, we developed a windowed multi‐attribute data input method that incorporates neighboring information to ensure lateral consistency. Based on the geological understanding that different attributes computed from the same sample convey correlated features reflecting geological properties, we pre‐train the network on large‐scale unlabeled data using a self‐supervised learning strategy of attribute masking and recovery. This approach encourages the network to learn correlation features, thereby improving its generalization by familiarizing it with the data distribution across the entire study field. We then employ the LoRA finetuning method to adapt the pre‐trained model to gas saturation prediction with meager labeled data while preserving pre‐trained knowledge. We applied this method to field data in the South China Sea, achieving accurate and generalized gas saturation predictions.https://doi.org/10.1029/2025JH000631self‐supervised learningfew‐shot problemgas saturation predictiondeep learningseismic attributes
spellingShingle Long Han
Xinming Wu
Renjie Chen
Yunhua Shi
Zhanxuan Hu
Huijing Fang
Self‐Supervised Pre‐Training and Few‐Shot Finetuning for Gas‐Bearing Prediction
Journal of Geophysical Research: Machine Learning and Computation
self‐supervised learning
few‐shot problem
gas saturation prediction
deep learning
seismic attributes
title Self‐Supervised Pre‐Training and Few‐Shot Finetuning for Gas‐Bearing Prediction
title_full Self‐Supervised Pre‐Training and Few‐Shot Finetuning for Gas‐Bearing Prediction
title_fullStr Self‐Supervised Pre‐Training and Few‐Shot Finetuning for Gas‐Bearing Prediction
title_full_unstemmed Self‐Supervised Pre‐Training and Few‐Shot Finetuning for Gas‐Bearing Prediction
title_short Self‐Supervised Pre‐Training and Few‐Shot Finetuning for Gas‐Bearing Prediction
title_sort self supervised pre training and few shot finetuning for gas bearing prediction
topic self‐supervised learning
few‐shot problem
gas saturation prediction
deep learning
seismic attributes
url https://doi.org/10.1029/2025JH000631
work_keys_str_mv AT longhan selfsupervisedpretrainingandfewshotfinetuningforgasbearingprediction
AT xinmingwu selfsupervisedpretrainingandfewshotfinetuningforgasbearingprediction
AT renjiechen selfsupervisedpretrainingandfewshotfinetuningforgasbearingprediction
AT yunhuashi selfsupervisedpretrainingandfewshotfinetuningforgasbearingprediction
AT zhanxuanhu selfsupervisedpretrainingandfewshotfinetuningforgasbearingprediction
AT huijingfang selfsupervisedpretrainingandfewshotfinetuningforgasbearingprediction