Search

Real-time drilling analysis requires knowledge of lithology at the drill bit. However, logging-while-drilling (LWD) sensors in the bottom hole assembly (BHA) are usually positioned 2–50 m (7–164 ft) above the bit (called the sensor offset), leading to a delay in real-time drilling analysis. The current industry solution to overcome this delay involves stopping drilling to perform a bottoms-up circulation for cuttings evaluation—a process that is both time-consuming and costly. To address this issue, our study evaluates three methodologies for real-time lithology prediction at the bit using drilling and petrophysical parameters. The first method employs a petrophysical approach, which involves using bulk density and neutron porosity predicted at the bit. The second method combines unsupervised and supervised machine learning (ML) for prediction. The third method employs classification algorithms on manually labeled lithology data from mud log reports, a novel approach used in this work. Our results show varying degrees of success: the bulk density versus neutron porosity cross-plot method achieved an accuracy of 58% with blind-well test data; the ML approach improved accuracy to 66%; and the Random Forest (RF) classification with manual labeling significantly increased accuracy to 86%. This comparative analysis of three different methodologies for lithology prediction has not been previously explored in the literature. While clustering and classification methods have been regarded as the most effective, our study demonstrates that they do not always yield the best result. These findings demonstrate that ML models, particularly the manual labeling approach, substantially outperform the petrophysical method. This new algorithm, designed for real-time applications, uses selected input parameters to effectively minimize problems associated with the sensor offset of LWD tools. It rapidly adapts to changes, offering a quicker and more cost-effective interpretation of lithology. This eliminates the need for time-consuming bottoms-up circulation to evaluate cuttings. Ultimately, this approach enhances drilling efficiency and significantly improves the accuracy of lithology prediction, notably in identifying interbedded geological layers. [ABSTRACT FROM AUTHOR]

The sand bodies of the delta front reservoirs in the WT Sag have small scales and strong heterogeneity, which leads to the difficulty of lithology prediction and characterization for this type of reservoirs. According to core analysis and logging data, muddy sandstone reservoirs are widely developed in the target layer of the study area, but the difference in compressional wave impedance between muddy sandstone, fine sandstone, and mudstone is small, and it is not possible to effectively use compressional wave impedance to identify the three lithologies. Moreover, the well network in the study area is sparse, and it is not possible to effectively extract the variogram, which affects the effect of post-stack seismic stochastic inversion and further affects the prediction of reservoir lithology. To solve these complex problems, the compressional wave impedance curve was reconstructed by using wavelet reconstruction and information statistics weighting methods. Sensitivity analysis of seismic attributes to sand to ground ratio was conducted, and root mean square (RMS) attributes were applied to characterize the sand body and extract a variation function from it. Random inversion of post stack seismic data was performed using reconstructed curves combined with variation functions. The results show that the compressional wave impedance reconstructed by information statistics weighting has a good identification effect on lithology. The lithology model established by post-stack seismic stochastic inversion and Bayesian algorithm matches well with the validation well lithology. The planar distribution of sand ratio made by the lithology model is consistent with geological understanding. It can be seen that the compressional wave impedance stochastic inversion model established by using the above methods has high reliability, and has important exploration guidance significance for predicting the planar distribution of sand bodies and designing horizontal well trajectories. [ABSTRACT FROM AUTHOR]

With the escalating demand for underground mining and infrastructure construction, the optimization of tunnel construction has emerged as a primary concern for researchers. The geological conditions encountered during the excavation of hard rock tunnels using tunnel boring machines (TBM) significantly impact construction efficiency and cost-effectiveness. The existing lithology testing methods need to be more efficient in aligning with TBM operational efficiency. In recent years, the rapid advancement of artificial intelligence has paved the way for its integration into numerous domains, including tunnel engineering. To address this issue, this study proposes three innovative hybrid RF-based intelligent models, namely PSO-RF, ALO-RF, and GWO-RF, for the precise prediction of lithology in hard rock tunnels using TBM working parameters. The TBM operating parameters of the Jilin Yinsong Water Supply Project serve as the basis for this investigation. Twelve distinct characteristic parameters relevant to the lithology of the tunnel working face were carefully selected as input parameters for lithology prediction. Comparative analysis of the three hybrid models reveals that GWO-RF demonstrates exceptional lithology prediction performance (ACC = 0.999924; PREA = 0.0.9999976; RECA = 0.999775; F1A = 0.999876; Kappa = 0.999911), whereas PSO-RF and ALO-RF exhibit slightly inferior performance. Nonetheless, all three hybrid models exhibit a significant improvement in prediction accuracy compared to the unoptimized RF model. The research findings presented herein facilitate the swift determination of TBM working surface lithology, enabling timely adjustment of TBM working parameters, reducing equipment wear and tear, and enhancing construction efficiency. [ABSTRACT FROM AUTHOR]

Identification and characterisation of reservoir facies is a prime factor in delimiting the hydrocarbon potential zones of a reservoir for hydrocarbon exploration. The geophysical logs, which are physical parameters of reservoir facies measured in the vicinity of boreholes, play a crucial role in the interpretation of reservoir facies. The present study deals with the identification of the lithology of the Limbodara oil field in the Cambay basin using machine learning (ML) techniques on geophysical logs. The supervised techniques of machine learning, such as support vector machines (SVM), artificial neural networks (ANN), and k-nearest neighbours (kNN), are used as nonlinear classifiers for the identification of lithology from nonlinear geophysical logs. The hyperparameters of the ML model are optimised using the grid search cross-validation (CV) method to increase the performance of the model, as evaluated by confusion matrix, area under receiver operating characteristics curve (AUC), precision, recall, and F1 score. The ML model used five geophysical parameters of two wells with four known distinguished lithologies (Class-A, Class-B, Class-C, and Class-D) for optimisation and training of the model. The optimised and trained model for each lithology for kNN, SVM, and ANN shows an overall correct prediction of true values with 85.4, 87.0, and 88.9%, respectively, from the confusion matrix. Apart from this, the receiver operative characteristics (ROC) also show that the overall area under the curve for each lithology is greater than 90%, and other evaluation parameters such as precision, recall, and F1 score show accuracy greater than 84%, except for the cases of Class C and Class D from SVM and ANN. Thus, the accuracy of each model from evaluation parameters suggests that the combined analysis of different ML models offers to select the optimised ML model for better results and validation to achieve and model the lithology with better precision. Highlights: A way out for obtaining litholog supplements at uncored section in boreholes Established ML assisted mapping function between wireline logs and lithologs Predicted litholog sequence with secure level of accuracy (>80%) [ABSTRACT FROM AUTHOR]

Petrographic prediction is crucial in identifying target areas and understanding reservoir lithology in oil and gas exploration. Traditional logging methods often rely on manual interpretation and experiential judgment, which can introduce subjectivity and constraints due to data quality and geological variability. To enhance the precision and efficacy of lithology prediction, this study employed a Savitzky–Golay filter with a symmetric window for anomaly data processing, coupled with a residual temporal convolutional network (ResTCN) model tasked with completing missing logging data segments. A comparative analysis against the support vector regression and random forest regression model revealed that the ResTCN achieves the smallest MAE, at 0.030, and the highest coefficient of determination, at 0.716, which are indicative of its proximity to the ground truth. These methodologies significantly enhance the quality of the training data. Subsequently, a Transformer–long short-term memory (T-LS) model was applied to identify and classify the lithology of unexplored wells. The input layer of the Transformer model follows an embedding-like principle for data preprocessing, while the encoding block encompasses multi-head attention, Add & Norm, and feedforward components, integrating the multi-head attention mechanism. The output layer interfaces with the LSTM layer through dropout. A performance evaluation of the T-LS model against established rocky prediction techniques such as logistic regression, k-nearest neighbor, and random forest demonstrated its superior identification and classification capabilities. Specifically, the T-LS model achieved a precision of 0.88 and a recall of 0.89 across nine distinct lithology features. A Shapley analysis of the T-LS model underscored the utility of amalgamating multiple logging data sources for lithology classification predictions. This advancement partially addresses the challenges associated with imprecise predictions and limited generalization abilities inherent in traditional machine learning and deep learning models applied to lithology identification, and it also helps to optimize oil and gas exploration and development strategies and improve the efficiency of resource extraction. [ABSTRACT FROM AUTHOR]

The accurate prediction of underground formation lithology class and tops is a critical challenge in the oil industry. This paper presents a machine-learning (ML) approach to predict lithology from drilling data, offering real-time litho-facies identification. The ML model, applied via the web app "GeoVision", achieves remarkable performance during its training phase with a mean accuracy of 95% and a precision of 98%. The model successfully predicts claystone, marl, and sandstone classes with high precision scores. Testing on new data yields an overall accuracy of 95%, providing valuable insights and setting a benchmark for future efforts. To address the limitations of current methodologies, such as time lags and lack of real-time data, we utilize drilling data as a unique endeavor to predict lithology. Our approach integrates nine drilling parameters, going beyond the narrow focus on the rate of penetration (ROP) often seen in previous research. The model was trained and evaluated using the open Volve field dataset, and careful data preprocessing was performed to reduce features, balance the sample distribution, and ensure an unbiased dataset. The innovative methodology demonstrates exceptional performance and offers substantial advantages for real-time geosteering. The accessibility of our models is enhanced through the user-friendly web app "GeoVision", enabling effective utilization by drilling engineers and marking a significant advancement in the field. [ABSTRACT FROM AUTHOR]

Machine Learning (ML) and Deep Learning (DL) have demonstrated notable success in diverse geoscience domains. However, the traditional ML/DL algorithms often rely on laborious trial-and-error processes to identify suitable hyperparameters and neural architectures, necessitating advanced ML expertise. The emerging paradigm of Automated Machine Learning (AutoML) and Automated Deep Learning (AutoDL) aims to generate high-quality ML/DL models with minimum human intervention. A typical AutoDL pipeline encompasses data preparation, feature engineering, model generation, and model evaluation. In this study, we explore the potential of the AutoDL plug-in for hyperparameter and architecture optimizations, to address three geoscience challenges: classification task of lithology prediction and regression tasks such as shear wave velocity (V s) and total organic carbon (TOC) predictions. Our results demonstrate promising outcomes for these tasks, potentially offering greater reliability and reproducibility than the ML algorithms. Furthermore, AutoDL presents two practical advantages: (1) Artificial Intelligence or AI democratization, which aspires to provide individuals with limited ML/DL expertise access to this approach's benefits, and (2) reduced carbon footprint by circumventing energy-intensive trial-and-error tasks associated with traditional ML/DL, thus promoting a more sustainable future for humanity. • AutoDL-Supported algorithms offer advantages over traditional ML/DL in Geosciences, automating workflows to save time and resources. • AutoDL improves tasks such as lithology, S-wave velocity, and TOC predictions, potentially surpassing conventional ML/DL. • AutoDL democratizes AI for all, fostering sustainability by reducing energy-intensive trial-and-error. [ABSTRACT FROM AUTHOR]

The common scientific issue presented by lithology prediction of tight sandstone reservoirs is that most primary lithologies cannot be distinguished effectively within the classic crossplots since they universally have similar logging responses. Lithology prediction actually is an issue of pattern recognition, and has been proved its best solver currently is machine learning technique. XGBoost is demonstrated to be a high-efficient predictor, while two factors, setting of hyper-parameters and dimensionality reduction of learning dataset, severely limit its computational performance. Genetic algorithm-particle swarm optimization (GA-PSO) and continuous restricted Boltzmann machine (CRBM) are introduced to optimize hyper-parameters and reduce amount of learning variables for the calculation of XGBoost, respectively, then a smart predictor CRBM-GA-PSO-CRBM proposed. Data used for validation is collected from the wells in the member of Chang 4 + 5, Jiyuan Oilfield, Ordos Basin. Three experiments are designed to verify prediction capability of the proposed model. Compared to other validated models, the higher prediction accuracies are all generated by the proposed model in three experiments, well expounding the proposed model is capable to produce reliable predicted lithologies, and has a better generalization and application prospect in the study field of lithology prediction. [ABSTRACT FROM AUTHOR]

Lithology prediction, especially for reservoirs consisting of complex lithologies, is universally considered as a critical underlying task for petroleum exploration, because lithological data is indispensable for the analysis of some geological work, such as stratigraphic correlation or sedimentation modeling. Hence, how to acquire the reliable lithological information gradually becomes a hot topic in the geoscience. Probabilistic neural network (PNN) is an excellent approach for lithology prediction since it can efficiently complete pattern recognition by determining characteristics of each kind of learning data. However, its computation performance is severely limited by two factors, which are quality of raw data and selection of parameter. High correlation of raw data could be an obstacle for the determination of data characteristics because partial probability density distributions established by PNN would be merged. As window length of each probability density distribution shows great impact on the accuracy of calculated results, the selection of this parameter must be addressed optimally before prediction. In order to improve the calculation capability of PNN, two techniques, continuous restricted Boltzmann machine (CRBM) and particle swarm optimization (PSO), are introduced. CRBM has the special function of extracting features from raw data and the features generally present with low correlation, thus can be viewed as an ideal preprocessing segment for PNN. PSO is one of the most efficient algorithms used for solving optimization problem, and the optimal parameter setting of PNN, therefore, can be revealed. Due to the advantages of CRBM and PSO for PNN, a new method for complex lithology prediction is proposed, which is referred as CRBM-PSO-PNN. Data for new method validation is recorded by two wells which are located in the IARA oilfield. Moreover, three experiments are well designed in order to verify the computing capability of new method comprehensively. Experiment results manifest that the prediction accuracies provided by new method in three experiments are highest, all of which are over 75%. High prediction accuracies fully demonstrate that the proposed method is effective to predict complex lithology, and the predicted results are reliable to serve other geological work. • PNN is potential to solve complex lithology prediction. • CRBM is capable to reduce correlation degree of learning data. • Optimal spread of each probability density distribution can be known by PSO. • Extension of learning dataset can improve prediction accuracy of CRBM-PSO-PNN. [ABSTRACT FROM AUTHOR]

Deltaic petroleum systems are often complex and heterogeneous. Delineation of sandstone bodies in such complex depositional settings constitutes a crucial challenge in reservoir exploration. However, reservoir characterization based on conventional well-logs and stratigraphic interpretation in complex settings often leads to uncertainties. In this work, we present an integrated quantitative interpretation to reveal heterogeneity and the reservoir properties in a complex deltaic environment. The study benefits from integrating borehole lithology logs, electrical well logging, and a 3D post-stack seismic cube. To demonstrate the practical application and effectiveness of our proposed approach, seismic inversion has been applied to the Acacus Formation in the Ghadames Basin of southern Tunisia. We estimate the acoustic impedance volume from 3D seismic data using a stochastic inversion algorithm. This algorithm enables the generation of multiple realizations, accounting for uncertainties and providing a more comprehensive understanding of subsurface impedance. Then, the porosity model is computed by establishing a functional relationship between acoustic impedance and porosity. The analysis of the relationship between acoustic impedance and different rock properties helped to differentiate lithological units of sandstone and claystone. This lithology classification is used to estimate the lithofacies model. Finally, the obtained porosity and lithofacies models are validated with well data. The crucial correlation between facies distribution and porosity allows for accurate mapping of individual thin sandstone bodies and their property distribution beyond well control. The lithofacies model shows significant sandstone bodies in the Acacus A and C units that constitute good reservoirs. The porosity model confirms this, as these sandstone facies exhibit high porosity (20%). The proposed approach has proven to be powerful in reservoir delineation, characterization, and exploration, and can be applied in similar geological setting basins, and frameworks. • Acoustic impedance volume is estimated from seismic data using a stochastic inversion algorithm. • Acoustic impedance's relationship with different rock properties is analyzed, for sandstone and claystone differentiation. • Porosity cube is obtained from a linear relationship with acoustic impedance. • Lithofacies model is derived from a lithological classification using acoustic impedance cross-plot versus VpVs ratio. • Heterogeneity and reservoir characteristics of the Acacus sandstone are revealed based on an integrated interpretation. [ABSTRACT FROM AUTHOR]

Abstract Technologies of real-time data measurement during drilling operation have kept the attention of petroleum industries in the past years, especially with the benefit of real-time formation evaluation through logging-while-drilling technology. It is expected that most of the logging data will be recorded in real-time operation. Hence, application of automated lithology prediction tool will be essential. An automatic method to predict lithology from borehole geophysical data was developed. It was solved as a multivariate classification problem with multidimensional explanatory variables. The learning algorithm combines kernel density estimates and a classification rule that is based on these estimates. The goal of this work is to test the method on a univariate variable and validate the prediction accuracy by calculating the misclassification rates. In addition, the results will be established as a baseline for application in practice and future developments for multivariate variables analysis. Gamma-ray from wireline logging is selected as the variable to describe two lithology groups of shale and not-shale. Data from six wells in the Norwegian Continental Shelf were extracted and examined with aids of explorative data analysis and hypothesis testing, and then divided into a training- and test data set. The selected algorithm processed the training data into models, and later each element of test data was assigned to the models to get the prediction. The results were validated with cutting data, and it was proved that the models predicted the lithology effectively with misclassification rates less than 15% at its lowest and average of ± 31%. Moreover, the results confirmed that the method has a promising prospect as lithology prediction tool, especially in real-time operation, because the non-parametric approach allows real-time modeling with fewer data assumptions required. Highlights • An automated lithology prediction method has been successfully developed. • Accurate lithology predictions were achieved with low misclassification rates. • The method is feasible to be applied in practice, especially in a real-time operation. [ABSTRACT FROM AUTHOR]

The upper Ordovician reservoir is one of the leading producing units in SW part of the Murzuq basin; it has a complex architecture inherited from the glacial effects on the braided fluvial deposits of the late stage at this era. The ultimate target for any petrophysical evaluation is to determine the quantities of water saturation and porosity accurately, but unfortunately, the models for thinly interbedded sand and shale layers are not straightforward. Therefore in this paper, an effort has been made to delineate and distinct between the types of interbedded (shaly -sand and sandyshale layers), and lithology end members by application compressional compliance versus density cross-plot in the two studied wells. After that, the determination of the thinly interbedded of layers type is augmented by correlating Vp-compliance with Poisson's ratio, effective porosity, and VP/VS. Among all the examined cross-plot; Poison's ratio, VP/VS and effective porosity with compressional compliance enhanced the upper and lower boundaries of the thinly interbedded sand and shale layers of the Ordovician succession in Murzuq basin. The "Mamuniyat" formation (main reservoir) with more clean sand content displays low compressional compliance, low-velocity ratio, and low Poisson's ratio. In contrary, the rich TOC shale (Hot shale) shows high compressional compliance, high-velocity ratio and high Poisson's ratio. While the disputed sandstone, siltstones and silty shales of late Ordovician "Bir Tlacsin" formation has slightly higher compressional compliance, velocity ratio, and Poisson's ratio than the underlying Mamuniyat formation, which enables drawing a clear contact between two gradually graded formations in the areas with no abrupt changes. The estimated petrophysical and elastic and properties are then used as an input for electrofacies prediction in both wells, by using unsupervised neural network classification. The predicted petrophysical facies clusters in both wells failed to differentiate between the various type of shales. However, the petro-elastic facies cluster reliably delineated the interbedded sandy- shale and shaly-sand thin layers without using gamma-ray logs. The results will help to avoid and to reduce the errors made during fluid substitution and rock physics models of shaly -sand formations. [ABSTRACT FROM AUTHOR]

Effective characterization of lithology is vital for the conceptualization of complex aquifer systems, which is a prerequisite for the development of reliable groundwater-flow and contaminant-transport models. However, such information is often limited for most groundwater basins. This study explores the usefulness and potential of a hybrid soft-computing framework; a traditional artificial neural network with gradient descent-momentum training (ANN-GDM) and a traditional genetic algorithm (GA) based ANN (ANN-GA) approach were developed and compared with a novel hybrid self-organizing map (SOM) based ANN (SOM-ANN-GA) method for the prediction of lithology at a basin scale. This framework is demonstrated through a case study involving a complex multi-layered aquifer system in India, where well-log sites were clustered on the basis of sand-layer frequencies; within each cluster, subsurface layers were reclassified into four depth classes based on the maximum drilling depth. ANN models for each depth class were developed using each of the three approaches. Of the three, the hybrid SOM-ANN-GA models were able to recognize incomplete geologic pattern more reasonably, followed by ANN-GA and ANN-GDM models. It is concluded that the hybrid soft-computing framework can serve as a promising tool for characterizing lithology in groundwater basins with missing lithologic patterns. [ABSTRACT FROM AUTHOR]

Lithology prediction is one the most affective requirements in all of the petroleum engineering embranchments. Petrophysical analysis, geophysical modeling, statistical methods and artificial intelligent approaches have been used to lithology prediction. Support vector machines (SVMs) based on statistical learning theory (SLT) and the principles of structural risk minimization (SRM) and empirical risk minimization (ERM) use an analytical approach to classification and regression. In this research, SVM classification method is used to lithology prediction from inverted seismic attributes data and petrophysical logs based on petrographic studies of cores lithology in a heterogeneous carbonate reservoir in Iran. Also, because of high impact of the data set size on most of machine learning techniques, effect of training data set size on different SVMs was deliberated by training and testing SVMs by six different partitioned cases according to the learning ratio of each case. Data preparation including normalization, attribute selection, kernel parameters optimization by grid search technique and data partitioning to construct training and testing data sets were performed on the data. The results showed that the SVM performs well in lithology prediction using inverted seismic attributes data and petrophysical logs, and by training data set size reduction, SVM performance has not affected too much, which it is an advantage for SVM as a machine learning method. Also, in order to predict lithology by SVMs using small training data sets, it is recommended to use normalized polynomial kernel function. Kernel functions and generally SVMs are not affected by the training data set size when the learning ratio varies in normal learning ratios. Using the kernels with their associated optimum values of the parameters obtained from grid search technique, it is possible to predict lithology in the investigated reservoir. [ABSTRACT FROM AUTHOR]

Lithology prediction is one of the most important issues in the petroleum geology and geological studies of petroleum engineering. Since well logging responses are very analogous for heterogeneous carbonate and evaporite sequences, a precisionist lithology prediction at predetermined depths becomes extremely critical. In this work, a combination of conventional petrophysical-based method and artificial intelligent approaches are used for lithological characterization of these layered reservoirs. Support vector machines (SVMs) are based on statistical learning theory and the principles of structural and empirical risk minimization use a non-heuristic analytical approach for prediction. SVM classification method is adopted for lithology prediction from petrophysical well logs based on core analysis data in an Iranian heterogeneous carbonate reservoir consisting of limestone, dolomite and anhydrite sequences. Normalization and attribute selection are conducted for data preparation purposes and the effect of kernel functions types on SVM performance is then investigated. Results show that SVM is a useful approach for lithology prediction and the radial basis function kernel is more accurate as compared to other kernel functions since it yields minimum misclassification rate error. [ABSTRACT FROM AUTHOR]

Soft computing tools play a vital role in fixing certain non-linear problems related to the earth. More specifically, digging out the mysteries of subsurface of the earth, the nonlinearity can be converging to assemble an approximate solution which resembles the real characteristics of the earth. Adaptive Neuro Fuzzy Inference System (ANFIS) tool is one of the best soft computing tools to estimate the complex data analysis. ANFIS was applied to estimate the subsurface parameters of earth using the Vertical Electrical Sounding (VES) data. Classifying the lithology based on the resistivity values by ANFIS is employed here in this paper. As the resistivity of each formation varies in range of values, ANFIS tool thus approximates the subsurface features based on effective training. In this study, ANFIS performance was checked with training data, and successively it has been tested with the field data. Optimized ANFIS algorithm provides the necessary tool for predicting the non-linear subsurface features. The best training performance of this soft computing tool efficiently predicts the subsurface lithology. Also the interpreted results show the true resistivity and thickness of the subsurface layers of the earth. The proposed technique was represented in Graphical User Interface (GUI), and the lithological variables are predicted in texture format and linguistic variables. [ABSTRACT FROM AUTHOR]

Lithologies are significant indicators to get deep insight of depositional and mineralogical properties of target formations, and the classic approach of achieving them is crossplot. Nonetheless, crossplot presents ineffectively when addressing classification of tight sandstone reservoirs, since most primary lithological components are characterized by similar logging responses. LightGBM (light gradient boosting machine) has been proved powerful to produce a remarkable classification, while its performance is seriously limited by the setting of hyper-parameters. LD-AFSA (linear decreasing-artificial fish swarm algorithm), an excellent solver for multi-objective optimization, then is introduced to modify the setting in a best circumstance. Besides, another integration for LightGBM is CRBM (continuous restricted Boltzmann machine), which specializes in generating less variables to speed up calculation. Consequently, a data-driven lithology predictor based on new ensemble learning is proposed, named CRBM-LD-AFSA-LightGBM. Data for validation of new predictor is cored by wells of Chang 4 + 5 member, Jiyuan Oilfield, Ordos Basin, and accordingly four experiments are designed to make a comprehensive evaluation. To highlight validating effect, SVM (support vector machine) and XGBoost (extreme gradient boosting) are adopted as competitors. Through comparison of experimental results, including prediction accuracy, F1-score, and AUC (area under curve), it is figured out that XGBoost-cored and LightGBM-cored predictors have capabilities to produce similar while more reliable results, meanwhile also exhibiting better generalization on prediction, but the computing time of latter predictor is only 1/25 shorter than that of the former. The results well demonstrate the proposed predictor plays a real high-efficient role in predicting lithologies and is deserved to receive a widespread employment in the field of logging interpretation because of its greater applicability. • Prediction capability of LightGBM can be enhanced by integration of CRBM and LD-AFSA. • CRBM-LD-AFSA-LightGBM is a real high-efficient predictor for lithologies. • Training more learning samples is effective to improve prediction effect. • A sparse learning data set can be processed by CRBM-LD-AFSA-LightGBM. [ABSTRACT FROM AUTHOR]

[Display omitted] • CRBM and PSO can be applied to improve the prediction efficiency of XGBoost. • CRBM-PSO-XGBoost is provably high-efficient for the lithology prediction. • Processing more learning samples is effective to enhance the prediction accuracy. • CRBM-PSO-XGBoost is proved to be robust even dealing with a sparse learning data. Acquiring reliable lithology information is a critical step for geological analysis since many basic jobs in early exploration have to be completed under the application of lithological materials. Lithology prediction then is always regarded as a research hotspot in geosciences. XGBoost is proved to be more powerful on pattern recognition than classic models, as it takes advantages of gradient boosting, classification tree, regularization, and other advanced machine learning techniques, thus being more potential to provide an ideal solution for lithology prediction. Nonetheless, this model is difficult to obtain optimal results due to the employment of many hyper-parameters, and will be low-efficient when dealing with many variables. Therefore, two computing techniques, continuous restricted Boltzmann machine (CRBM) and particle swarm optimization (PSO), are introduced to improve prediction performance of XGBoost. CRBM can extract fewer while more significant features from original data, and PSO will automatically optimize hyper-parameters during training process. Data used for validation is derived from tight sandstone reservoirs of member of Chang 4 + 5, western Jiyuan Oilfield, Ordos Basin, Northern China. Three experiments are designed to verify prediction capability of the proposed model. In order to highlight validation effect, two classic predictors named support vector machine (SVM) and gradient boosting decision tree (GBDT) are applied to create a contrast. The total prediction accuracy and the respective accuracy of each lithology produced by CRBM-PSO-XGBoost are all the highest in three experiments, well demonstrating the proposed model is effective to predict the lithology of tight sandstone reservoirs and has better robustness. [ABSTRACT FROM AUTHOR]

The identification of lithology while drilling with horizontal well bit is a difficult problem to solve in geosteering. However, due to the existence of "zero length" (the distance between the logging tool and the bit), the lithology at the horizontal well bit cannot be accurately interpreted in real time, which creates a lag in geosteering. Based on logging while drilling (LWD) technology, this paper uses supervised learning in machine learning methods, conventional logging interpretation technology and big data idea in modern computer science, through interdisciplinary theories and methods to research lithology prediction for horizontal well bits in reservoirs. First, a measurement point and vertical reservoir boundary distance (D-MP-VRB) database is built according to different step size categories. Second, based on the D-MP-VRB database, D-MP-VRB prediction models are established using one-versus-one support vector machines (OVO SVMs), random forest (RF), neural networks (NN), and extreme gradient boosting tree (XGBoost) algorithms. To reduce the dimensions of the input data, the feature parameters of the samples are obtained by a correlation analysis of the logging data. The optimal parameter values of each algorithm are determined by grid search and 10-fold cross-validation methods. Finally, the prediction formula of the bit and vertical reservoir boundary distance based on the D-MP-VRB prediction model is established. A case study is performed with data from a sandstone reservoir in Changqing oilfield, Ordos Basin, China. On this basis, the lithology predictions at the bit in real time are carried out by using four models. Considering the principle of model prediction accuracy, through 1320 experiments, only the XGBoost prediction model can be selected, and the step size of the target category is 2 m, however, this model takes the longest time to train. Therefore, the reliable prediction model trained by the sample data of the original training set is used to predict the reservoir information encountered by horizontal well. After the newly drilled reservoir information has accumulated to a certain amount and accurately explained, it is added to the original training set sample data, and the prediction model is retrained to improve the accuracy and adaptability of the model. Based on the prediction results of the XGBoost model and the prediction formula, the distance prediction between the horizontal well bit and the vertical reservoir boundary is realized, real-time lithology correction at the bit is realized, and the adverse effect of the "zero length" on the lithology prediction at the bit is reduced. The research results provide not only a new method for the real-time prediction of lithology in horizontal well bits but also a theoretical basis for the geosteering of oilfield development and valuable information for future research. • A new method for real-time prediction of measurement point and vertical reservoir boundary distance (D-MP-VRB) while drilling is proposed. Real-time prediction models for D-MP-VRB were established and optimized by combining LWD technology with machine learning method based on OVO SVMs, RF, NN and XGBoost algorithms. The simultaneous of efficient drilling and D-MP-VRB prediction is realized. • A prediction formula of the horizontal well bit and vertical reservoir boundary distance based on the D-MP-VRB prediction model is established. The distance prediction between the horizontal well bit and the vertical reservoir boundary is realized, real-time lithology correction at the bit is realized, and the adverse effect of "zero length" on the lithology prediction at the bit is reduced. [ABSTRACT FROM AUTHOR]