Search

In the field of reservoir characterization and management, the completeness and accuracy of geophysical logs are pivotal. Often, these logs are marred by missing segments or distortions due to logistical and environmental challenges in boreholes. To address this issue, our study introduces synergistic method combining log data preconditioning with advanced machine learning (ML) techniques—including k-nearest neighbors, support vector machine, decision tree, random forest, extreme gradient boosting, gaussian process regression, and artificial neural networks. We focus on uncovering the complex, nonlinear relationships inherent in geophysical logs through a robust analysis involving a correlation matrix and F-test for predictor significance and ranking. We applied this approach to the wireline logs of the Lakadong-Therria Formation in the Bhogpara oil field, India, to demonstrate the effectiveness of ML in reliably predicting missing logs. Notably, our ML models adeptly forecast bulk density logs using data from gamma-ray, deep resistivity, neutron porosity, and photoelectric factor logs. The high R-squared correlation coefficients achieved (R2 score: over 0.89 in training and 0.85 in testing phases) attest to the accuracy of our predictions. This study proves the efficacy of ML in enhancing reservoir analysis and offers a cost-effective tool for accurately reconstructing any geophysical log, paving the way for more informed decisions in reservoir management. [ABSTRACT FROM AUTHOR]

This study applies machine learning techniques to improve petrographic classification in India's Bokaro coalfield's Barakar Formation, using conventional geophysical well logs from three wells. We analysed natural gamma ray, true resistivity, bulk density, neutron porosity, and photoelectric factor data using k-nearest neighbor (kNN), support vector machine (SVM) and random forest (RF) classifiers. A master well provided initial reference log measurement cut-off values for typical lithologies like shale, sandstone, carbonaceous shale, and coal, forming the basis of our training dataset. We assessed model accuracy using precision, recall, and F1-score metrics, finding the random forest model to be the most effective in litho-type discrimination. During the training phase, the computed overall accuracy of the predicted ML modes exceeded 89% and model accuracy hierarchy was RF>SVM>kNN. These classifiers were then applied to other well locations to predict lithological sequences, aiding in lithofacies sequence identification and potential fault extension detection. The study demonstrates the random forest model's superior precision and efficiency in lithological discrimination. Our findings enhance automated processes for identifying missing lithology during well correlation, offering valuable insights for geological interpretation in resource exploration and development. This machine learning-driven approach marks a significant advancement in subsurface geological studies.Graphical abstract: This study applies machine learning techniques to improve petrographic classification in India's Bokaro coalfield's Barakar Formation, using conventional geophysical well logs from three wells. We analysed natural gamma ray, true resistivity, bulk density, neutron porosity, and photoelectric factor data using k-nearest neighbor (kNN), support vector machine (SVM) and random forest (RF) classifiers. A master well provided initial reference log measurement cut-off values for typical lithologies like shale, sandstone, carbonaceous shale, and coal, forming the basis of our training dataset. We assessed model accuracy using precision, recall, and F1-score metrics, finding the random forest model to be the most effective in litho-type discrimination. During the training phase, the computed overall accuracy of the predicted ML modes exceeded 89% and model accuracy hierarchy was RF>SVM>kNN. These classifiers were then applied to other well locations to predict lithological sequences, aiding in lithofacies sequence identification and potential fault extension detection. The study demonstrates the random forest model's superior precision and efficiency in lithological discrimination. Our findings enhance automated processes for identifying missing lithology during well correlation, offering valuable insights for geological interpretation in resource exploration and development. This machine learning-driven approach marks a significant advancement in subsurface geological studies. [ABSTRACT FROM AUTHOR]

In the E&P industry, accurate lithology classification is an essential task for successful exploration and production. Geophysical logs provide high-resolution petrophysical properties, but core logging is expensive and traditional techniques may not accurately classify lithologies. We demonstrated a comparative analysis of six ML algorithms: k-nearest neighbor (kNN), support vector machine (SVM), decision tree (DT), random forest (RF), extreme gradient boosting (XGBoost) and artificial neural network (ANN) for the prediction of lithologies from geophysical logs. Here we analysed the wireline logs of eight wells associated with the petroliferous Lakadong-Therria formation of the Bhogpara oil field of the Assam-Arakan Basin. This formation contains eight typical lithologies, namely clay stone, sand stone, calcareous sandstone, shale, calcareous shale, carbonaceous shale, coal and limestone. Performance of the ML algorithms were evaluated through accuracy, precision, recall, F1-score and receiver operating characteristic (ROC) curve. During the training and test phases, the computed overall accuracy of the predicted ML modes exceeded 82% and 71%, respectively. The model accuracy hierarchy was ANN > XGBoost > RF > SVM > DT > kNN during training, and ANN/XGBoost > kNN > DT/RF > SVM during testing. This approach allows interpreters to select the most accurate ML model based on training phase performance. This study provided a clear insight towards generating a supplement for litholog sequence and improving the accuracy and efficiency of lithology prediction in a geologically complex petroleum reservoir using pre-received core derived litholog information at few wells. [ABSTRACT FROM AUTHOR]

[Display omitted] • We exhibit state-of-the-art of geodetic velocity prediction using machine learning. • We forecast geodetic velocity in and around the Tibetan Plateau. • Geodetic velocities predicted using machine learning are extremely precise. • The demonstrated prediction approach saves both time and computation cost. So far only a few individuals have attempted to use Machine Learning approaches to anticipate GPS site velocity for crustal deformation research. Generally, a dense network of Continuously Operating Reference Station (CORS) is employed to monitor crustal deformation continuously. Campaign-mode GPS surveys are often used to densify the existing CORS network. Even sometimes, it is very challenging to establish a station in the location of our interest due to logistical problems and regional geographical considerations. However, this process is expensive, and crustal movement studies are eventually hampered by data missing issues owing to logistical restrictions. Thus, to obtain velocity vectors at the desired locations, we implemented machine learning (ML) techniques, such as support vector machines, decision trees and Gaussian processing regression, to accomplish the crustal movement precisely. We inspected data from 1271 permanent continuous and campaign-mode GPS stations, located on the Tibetan plateau and its environs. This study demonstrates the effectiveness of these ML techniques in forecasting velocity vectors (easting velocity (V E) and northing velocity (V N)) and enhancing plate movement characterisation. The correlation coefficient between (C R > 0.98 at training and C R > 0.96 at test phase) predicted and actual velocity vectors are satisfactory, making these ML predictive models considerably reliable for estimating geodetic velocity vectors. This ML algorithm demonstrates a remarkable achievement in the field of geodetic study in a cost-effective manner. [ABSTRACT FROM AUTHOR]

Rising greenhouse gas emissions, especially CO 2 , intensify global warming. Carbon Capture and Storage (CCS) is vital for reducing emissions from industrial sources, crucial for addressing climate change urgently. This study explores reservoir characterization challenges for CCS in the Penobscot field, offshore Nova Scotia. We used to cross-plots of geophysical logs to establish relationships between petrophysical properties and subsurface litho-facies, differentiating sand and shale facies within the Mississauga Formation (Early to Middle Cretaceous). Seismic impedance inversion distinguished between reservoir and non-reservoir facies. Multi-attribute assisted transformed seismic data and a support vector machine algorithm predicted litho-facies probabilities and effective porosity volumes in 3D space. The permutation predictor importance analysis was performed to assess the relative importance of input features in facies probabilities and effective porosity prediction. The structural component delineated the CO 2 injection trap boundary. Integrated analysis of structural configuration, litho-facies probability, effective porosity, and cap-rock sealing integrity validated zones for CO 2 storage. The dry, abandoned well L-30 in the structural high zone is recommended as a pilot CO 2 injector well. This study provides critical insights into reservoir characterization for CO 2 sequestration, aiding climate change mitigation through effective CCS, and recommends geomechanical studies and time-lapse seismic monitoring. • Identified promising sites for CO 2 storage in Penobscot field, Canada Offshore. • Predicted impedance, facies probabilities, and effective porosity with high accuracy. • Enhanced discrimination between the reservoir and non-reservoir facies. • Assessed lithological and petrophysical characteristics of the reservoir and cap rocks. • Emphasised storage site identification as a strategy to mitigate climate change. [ABSTRACT FROM AUTHOR]