To improve the accuracy and generalization of well logging curve reconstruction,this paper proposes an artificial intelligence large language model“Gaia”and conducts model evaluation experiments.By fine-tuning the p...To improve the accuracy and generalization of well logging curve reconstruction,this paper proposes an artificial intelligence large language model“Gaia”and conducts model evaluation experiments.By fine-tuning the pre-trained large language model,the Gaia significantly improved its ability in extracting sequential patterns and spatial features from well-log curves.Leveraging the adapter method for fine-tuning,this model required training only about 1/70 of its original parameters,greatly improving training efficiency.Comparative experiments,ablation experiments,and generalization experiments were designed and conducted using well-log data from 250 wells.In the comparative experiment,the Gaia model was benchmarked against cutting-edge small deep learning models and conventional large language models,demonstrating that the Gaia model reduced the mean absolute error(MAE)by at least 20%.In the ablation experiments,the synergistic effect of the Gaia model's multiple components was validated,with its MAE being at least 30%lower than that of single-component models.In the generalization experiments,the superior performance of the Gaia model in blind-well predictions was further confirmed.Compared to traditional models,the Gaia model is significantly superior in accuracy and generalization for logging curve reconstruction,fully showcasing the potential of large language models in the field of well-logging.This provides a new approach for future intelligent logging data processing.展开更多
Pore structure reflected from capillary pressure curves plays an important role in low-permeability formation evaluation. It is a common way to construct capillary pressure curves by Nuclear Magnetic Resonance（NMR） ...Pore structure reflected from capillary pressure curves plays an important role in low-permeability formation evaluation. It is a common way to construct capillary pressure curves by Nuclear Magnetic Resonance（NMR） log. However, the method＇s efficiency will be severely affected if there is no NMR log data or it cannot reflect pore structure well. Therefore, on the basis of J function and diagenetic facies classification, a new empirical model for constructing capillary pressure curves from conventional logs is proposed here as a solution to the problem. This model includes porosity and the relative value of natural gamma rays as independent variables and the saturation of mercury injection as a dependent variable. According to the 51 core experimental data sets of three diagenetic facies from the bottom of the Upper Triassic in the western Ordos Basin, China, the model＇s parameters in each diagenetic facies are calibrated. Both self-checking and extrapolation tests show a positive effect, which demonstrates the high reliability of the proposed capillary pressure curve construction model. Based on the constructed capillary pressure curves, NMR T_2 spectra under fully brine-saturated conditions are mapped by a piecewise power function. A field study is then presented. Agreement can be seen between the mapped NMR T_2 spectra and the MRIL-Plog data in the location of the major peak, right boundary, distribution characteristics and T_2 logarithmic mean value. In addition, the capillary pressure curve construction model proposed in this paper is not affected by special log data or formation condition. It is of great importance in evaluating pore structure, predicting oil production and identifying oil layers through NMR log data in low-permeability sandstones.展开更多
Well logging curves serve as indicators of strata attribute changes and are frequently utilized for stratigraphic analysis and comparison.Deep learning,known for its robust feature extraction capabilities,has seen con...Well logging curves serve as indicators of strata attribute changes and are frequently utilized for stratigraphic analysis and comparison.Deep learning,known for its robust feature extraction capabilities,has seen continuous adoption by scholars in the realm of well logging stratigraphic correlation tasks.Nonetheless,current deep learning algorithms often struggle to accurately capture feature changes occurring at layer boundaries within the curves.Moreover,when faced with data imbalance issues,neural networks encounter challenges in accurately modeling the one-hot encoded curve stratification positions,resulting in significant deviations between predicted and actual stratification positions.Addressing these challenges,this study proposes a novel well logging curve stratigraphic comparison algorithm based on uniformly distributed soft labels.In the training phase,a label smoothing loss function is introduced to comprehensively account for the substantial loss stemming from data imbalance and to consider the similarity between diff erent layer data.Concurrently,spatial attention and channel attention mechanisms are incorporated into the shallow and deep encoder stages of U²-Net,respectively,to better focus on changes in stratification positions.During the prediction phase,an optimized confidence threshold algorithm is proposed to constrain stratification results and solve the problem of reduced prediction accuracy because of occasional layer repetition.The proposed method is applied to real-world well logging data in oil fields.Quantitative evaluation results demonstrate that within error ranges of 1,2,and 3 m,the accuracy of well logging curve stratigraphic division reaches 87.27%,92.68%,and 95.08%,respectively,thus validating the eff ectiveness of the algorithm presented in this paper.展开更多
This paper introduces horizon control, seismic control, logging control and facies control methods through the application of the least squares fitting of logging curves, seismic inversion and facies-controlled techni...This paper introduces horizon control, seismic control, logging control and facies control methods through the application of the least squares fitting of logging curves, seismic inversion and facies-controlled techniques. Based on the microgeology and thin section analyses, the lithology, lithofacies and periods of the Permian igneous rocks are described in detail. The seismic inversion and facies-controlled techniques were used to find the distribution characteristics of the igneous rocks and the 3D velocity volume. The least squares fitting of the logging curves overcome the problem that the work area is short of density logging data. Through analysis of thin sections, the lithofacies can be classified into eruption airfall subfacies, eruption pyroclastic flow subfacies and eruption facies.展开更多
Derivative/volatility well-log attributes from very few commonly recorded well logs can assist in the prediction of total organic carbon(TOC)in shales and tight formations.This is of value where only limited suites of...Derivative/volatility well-log attributes from very few commonly recorded well logs can assist in the prediction of total organic carbon(TOC)in shales and tight formations.This is of value where only limited suites of well logs are recorded,and few laboratory measurements of TOC are conducted on rock samples.Data from two Lower-Barnett-Shale(LBS)wells(USA),including well logs and core analysis is considered.It demonstrates how well-log attributes can be exploited with machine learning(ML)to generate accurate TOC predictions.Six attributes are calculated for gamma-ray(GR),bulk-density(PB)and compressional-sonic(DT)logs.Used in combination with just one of those recorded logs,those attributes deliver more accurate TOC predictions with ML models than using all three recorded logs.When used in combination with two or three of the recorded logs,the attributes generate TOC prediction accuracy comparable with ML models using five recorded well logs.Multi-K-fold-cross-validation analysis reveals that the K-nearest-neighbour algorithm yields the most accurate TOC predictions for the LBS dataset.The extreme-gradient-boosting(XGB)algorithm also performs well.XGB is able to provide information about the relative importance of each well-log attribute used as an input variable.This facilitates feature selection making it possible to reduce the number of attributes required to generate accurate TOC predictions from just two or three recorded well logs.展开更多
基金Supported by the National Natural Science Foundation of China(52288101)National Key R&D Program of China(2024YFF1500600)。
文摘To improve the accuracy and generalization of well logging curve reconstruction,this paper proposes an artificial intelligence large language model“Gaia”and conducts model evaluation experiments.By fine-tuning the pre-trained large language model,the Gaia significantly improved its ability in extracting sequential patterns and spatial features from well-log curves.Leveraging the adapter method for fine-tuning,this model required training only about 1/70 of its original parameters,greatly improving training efficiency.Comparative experiments,ablation experiments,and generalization experiments were designed and conducted using well-log data from 250 wells.In the comparative experiment,the Gaia model was benchmarked against cutting-edge small deep learning models and conventional large language models,demonstrating that the Gaia model reduced the mean absolute error(MAE)by at least 20%.In the ablation experiments,the synergistic effect of the Gaia model's multiple components was validated,with its MAE being at least 30%lower than that of single-component models.In the generalization experiments,the superior performance of the Gaia model in blind-well predictions was further confirmed.Compared to traditional models,the Gaia model is significantly superior in accuracy and generalization for logging curve reconstruction,fully showcasing the potential of large language models in the field of well-logging.This provides a new approach for future intelligent logging data processing.
基金supported by the Scientific Research Starting Foundation of China University of Petroleum-Beijing at Karamay(No.RCYJ2016B-01-008)the Major National Oil&Gas Specific Project of China(No.2016ZX05050008)
文摘Pore structure reflected from capillary pressure curves plays an important role in low-permeability formation evaluation. It is a common way to construct capillary pressure curves by Nuclear Magnetic Resonance（NMR） log. However, the method＇s efficiency will be severely affected if there is no NMR log data or it cannot reflect pore structure well. Therefore, on the basis of J function and diagenetic facies classification, a new empirical model for constructing capillary pressure curves from conventional logs is proposed here as a solution to the problem. This model includes porosity and the relative value of natural gamma rays as independent variables and the saturation of mercury injection as a dependent variable. According to the 51 core experimental data sets of three diagenetic facies from the bottom of the Upper Triassic in the western Ordos Basin, China, the model＇s parameters in each diagenetic facies are calibrated. Both self-checking and extrapolation tests show a positive effect, which demonstrates the high reliability of the proposed capillary pressure curve construction model. Based on the constructed capillary pressure curves, NMR T_2 spectra under fully brine-saturated conditions are mapped by a piecewise power function. A field study is then presented. Agreement can be seen between the mapped NMR T_2 spectra and the MRIL-Plog data in the location of the major peak, right boundary, distribution characteristics and T_2 logarithmic mean value. In addition, the capillary pressure curve construction model proposed in this paper is not affected by special log data or formation condition. It is of great importance in evaluating pore structure, predicting oil production and identifying oil layers through NMR log data in low-permeability sandstones.
基金supported by the CNPC Advanced Fundamental Research Projects(No.2023ycq06).
文摘Well logging curves serve as indicators of strata attribute changes and are frequently utilized for stratigraphic analysis and comparison.Deep learning,known for its robust feature extraction capabilities,has seen continuous adoption by scholars in the realm of well logging stratigraphic correlation tasks.Nonetheless,current deep learning algorithms often struggle to accurately capture feature changes occurring at layer boundaries within the curves.Moreover,when faced with data imbalance issues,neural networks encounter challenges in accurately modeling the one-hot encoded curve stratification positions,resulting in significant deviations between predicted and actual stratification positions.Addressing these challenges,this study proposes a novel well logging curve stratigraphic comparison algorithm based on uniformly distributed soft labels.In the training phase,a label smoothing loss function is introduced to comprehensively account for the substantial loss stemming from data imbalance and to consider the similarity between diff erent layer data.Concurrently,spatial attention and channel attention mechanisms are incorporated into the shallow and deep encoder stages of U²-Net,respectively,to better focus on changes in stratification positions.During the prediction phase,an optimized confidence threshold algorithm is proposed to constrain stratification results and solve the problem of reduced prediction accuracy because of occasional layer repetition.The proposed method is applied to real-world well logging data in oil fields.Quantitative evaluation results demonstrate that within error ranges of 1,2,and 3 m,the accuracy of well logging curve stratigraphic division reaches 87.27%,92.68%,and 95.08%,respectively,thus validating the eff ectiveness of the algorithm presented in this paper.
基金A Project Funded by National Science and Technology Major Project (2011ZX05001-002-003)the Priority Academic Program Development of Jiangsu Higher Education Institutions (PAPD)Key Laboratory for Coalbed Methane Resources and Reservoir formation Process, CUMT, Ministry of Education, China
文摘This paper introduces horizon control, seismic control, logging control and facies control methods through the application of the least squares fitting of logging curves, seismic inversion and facies-controlled techniques. Based on the microgeology and thin section analyses, the lithology, lithofacies and periods of the Permian igneous rocks are described in detail. The seismic inversion and facies-controlled techniques were used to find the distribution characteristics of the igneous rocks and the 3D velocity volume. The least squares fitting of the logging curves overcome the problem that the work area is short of density logging data. Through analysis of thin sections, the lithofacies can be classified into eruption airfall subfacies, eruption pyroclastic flow subfacies and eruption facies.
文摘Derivative/volatility well-log attributes from very few commonly recorded well logs can assist in the prediction of total organic carbon(TOC)in shales and tight formations.This is of value where only limited suites of well logs are recorded,and few laboratory measurements of TOC are conducted on rock samples.Data from two Lower-Barnett-Shale(LBS)wells(USA),including well logs and core analysis is considered.It demonstrates how well-log attributes can be exploited with machine learning(ML)to generate accurate TOC predictions.Six attributes are calculated for gamma-ray(GR),bulk-density(PB)and compressional-sonic(DT)logs.Used in combination with just one of those recorded logs,those attributes deliver more accurate TOC predictions with ML models than using all three recorded logs.When used in combination with two or three of the recorded logs,the attributes generate TOC prediction accuracy comparable with ML models using five recorded well logs.Multi-K-fold-cross-validation analysis reveals that the K-nearest-neighbour algorithm yields the most accurate TOC predictions for the LBS dataset.The extreme-gradient-boosting(XGB)algorithm also performs well.XGB is able to provide information about the relative importance of each well-log attribute used as an input variable.This facilitates feature selection making it possible to reduce the number of attributes required to generate accurate TOC predictions from just two or three recorded well logs.
