No full-text available
To read the full-text of this research,
you can request a copy directly from the authors.
AI-Based Estimation of Hydraulic Fracturing Effect
Abstract
We studied the applicability of a gradient-boostingmachine-learning (ML) algorithm for forecasting of oil and total liquid production after hydraulic fracturing (HF). A thorough raw data study with data preprocessing algorithms was provided. The data set included 10 oil fields with more than 2,000 HF events. Each event has been characterized by well coordinates, geology, transport and storage properties, depths, and oil/liquid rates before fracturing for target and neighboring wells. Each ML model has been trained to predict monthly production rates right after fracturing and when the flows are stabilized. The gradient-boosting method justified its choice with R2 being approximately 0.7 to 0.8 on the test set for oil/total liquid production after HF. The developed ML prediction model does not require preliminary numerical simulations of a future HF design. The applied algorithm could be used as a new approach for HF candidate selection based on the real-time state of the field.
... A number of studies confirmed the effectiveness of ML models in application to oil recovery factor estimation [14][15][16]. Several other studies have reported the successful application of ML models to estimate the effects of hydraulic fracturing [17][18][19][20]. Kornkosky et al. applied multivariate linear regression to estimate the waterflooding effect [11]. ...
... It proves itself to be robust to noise, immune to multicollinearity, and sufficiently accurate for engineering applications [28]. The selected models are currently the most popular for similar regression problems [11,17,19,[29][30][31]. ...
... Nowadays, many IOR/EOR projects are being carried out worldwide. There are already examples of successful ML applications in the literature for hydraulic fracturing [17][18][19]. For such projects, it is crucial to assess the potential and risks in advance; however, this is not easy to do. ...
Waterflooding is a widely used secondary oil recovery technique. The oil and gas industry uses a complex reservoir numerical simulation and reservoir engineering analysis to forecast production curves from waterflooding projects. The application of such standard methods at the stage of assessing the potential of a huge number of projects could be computationally inefficient and requires a lot of effort. This paper demonstrates the applicability of machine learning to rate the outcome of waterflooding applied to an oil reservoir. We also explore the relationship of project evaluations by operators at the final stages with several performance metrics for forecasting. Real data about several thousand waterflooding projects in Texas are used in the current study. We compare the ML models rankings of the waterflooding efficiency and the expert rankings. Linear regression models along with neural networks and gradient boosting on decision threes are considered. We show that machine learning models allow reducing computational complexity and can be useful for rating the reservoirs, with respect to the effectiveness of waterflooding.
... Here it plays a key role to predict the production of the fractured CBM wells before the fracturing operation. There are many productivityprediction models for CBM wells, including the seepage mechanism-based methods [8][9][10][11] and the data-based machine learning methods [6,7]. ...
... In recent years, along with the development of big data techniques, machine learning has drawn much attention, such as multiple regression [8], random forest [9], support vector machine (SVM) and gradient-boosting [11] etc. However, their practical performance in field application are not satisfying enough. ...
... There are many researches focus on forecasting production capability of CBM wells and finding its controlling factors. The methods can be classified into 3 categories, the methods based on correlation [16][17][18][19], methods with interpretability [8][9][10][11], and with causality [32][33][34][35][36][37]. The Methods based on Correlation. ...
Machine learning approaches are widely studied in the production prediction of CBM wells after hydraulic fracturing, but merely used in practice due to the low generalization ability and the lack of interpretability. A novel methodology is proposed in this article to discover the latent causality from observed data, which is aimed at finding an indirect way to interpret the machine learning results. Based on the theory of causal discovery, a causal graph is derived with explicit input, output, treatment and confounding variables. Then, SHAP is employed to analyze the influence of the factors on the production capability, which indirectly interprets the machine learning models. The proposed method can capture the underlying nonlinear relationship between the factors and the output, which remedies the limitation of the traditional machine learning routines based on the correlation analysis of factors. The experiment on the data of CBM shows that the detected relationship between the production and the geological/engineering factors by the presented method, is coincident with the actual physical mechanism. Meanwhile, compared with traditional methods, the interpretable machine learning models have better performance in forecasting production capability, averaging 20% improvement in accuracy.
... Similarly, Al-Sudani et al. (2017) [10] introduced a control engineering system for real-time monitoring of drilling mechanical energy and bit wear, optimizing drilling performance. Meanwhile, for fracturing, Erofeev et al. (2021) [11] predicted post-hydraulic fracturing oil and liquid production with 80% accuracy, enabling real-time HF candidate selection. In the domain of oil recovery, Ouadi et al. (2023) [12] introduced high-accuracy models for predicting gas well productivity using Fishbone drilling, demonstrating its potential to enhance hydrocarbon recovery and reduce environmental impact. ...
... Similarly, Al-Sudani et al. (2017) [10] introduced a control engineering system for real-time monitoring of drilling mechanical energy and bit wear, optimizing drilling performance. Meanwhile, for fracturing, Erofeev et al. (2021) [11] predicted post-hydraulic fracturing oil and liquid production with 80% accuracy, enabling real-time HF candidate selection. In the domain of oil recovery, Ouadi et al. (2023) [12] introduced high-accuracy models for predicting gas well productivity using Fishbone drilling, demonstrating its potential to enhance hydrocarbon recovery and reduce environmental impact. ...
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.
... Having data on production prior to refracturing makes the production forecast problem easier to solve, compared to the case of production forecast after primary fracturing operations. This has also been noted in other studies [10]. ...
... These analogue wells search is very useful for a petroleum engineer as it allows to analyse fracturing operations, conducted previously, and the design parameters values, check whether an operation was successful or not, etc. We can also extract additional features from the neighbouring wells, which increase predictive power of the models [10]. For example, in our work, we used features, such as average fluid production divided by distance from the pilot well, using wells within 1 km from the pilot one. ...
We describe a stacked model for predicting the cumulative fluid production for an oil well with a multistage-fracture completion based on a combination of Ridge Regression and CatBoost algorithms. The model is developed based on an extended digital field data base of reservoir, well and fracturing design parameters. The database now includes more than 5000 wells from 23 oilfields of Western Siberia (Russia), with 6687 fracturing operations in total. Starting with 387 parameters characterizing each well, including construction, reservoir properties, fracturing design features and production, we end up with 38 key parameters used as input features for each well in the model training process. The model demonstrates physically explainable dependencies plots of the target on the design parameters (number of stages, proppant mass, average and final proppant concentrations and fluid rate). We developed a set of methods including those based on the use of Euclidean distance and clustering techniques to perform similar (offset) wells search, which is useful for a field engineer to analyze earlier fracturing treatments on similar wells. These approaches are also adapted for obtaining the optimization parameters boundaries for the particular pilot well, as part of the field testing campaign of the methodology. An inverse problem (selecting an optimum set of fracturing design parameters to maximize production) is formulated as optimizing a high dimensional black box approximation function constrained by boundaries and solved with four different optimization methods: surrogate-based optimization, sequential least squares programming, particle swarm optimization and differential evolution. A recommendation system containing all the above methods is designed to advise a production stimulation engineer on an optimized fracturing design.
... Deep learning is a popular research area in machine learning, which involves using large amounts of data to learn low-level features through deep structural learning and map them to high-level features in order to complete complex classification tasks by utilizing internal patterns in the data. In recent years, deep learning theory has continuously developed, and many scholars have begun to study the use of deep neural networks to process reservoir fracture data, which has also shown excellent performance in complex well-logging data processing applications [9][10][11][12][13]. Convolutional neural networks (CNN) have the ability to extract features and generalize [14,15] and can explore the nonlinear relationship between acoustic logging data and fracturing effect prediction without human intervention, making fracturing effect evaluation more objective [16]. ...
The utilization of hydraulic fracturing technology is indispensable for unlocking the potential of tight oil and gas reservoirs. Understanding and accurately evaluating the impact of fracturing is pivotal in maximizing oil and gas production and optimizing wellbore performance. Currently, evaluation methods based on acoustic logging, such as orthogonal dipole anisotropy and radial tomography imaging, are widely used. However, when the fractures generated by hydraulic fracturing form a network-like pattern, orthogonal dipole anisotropy fails to accurately assess the fracturing effects. Radial tomography imaging can address this issue, but it is challenged by high manpower and time costs. This study aims to develop a more efficient and accurate method for evaluating fracturing effects in tight reservoirs using deep learning techniques. Specifically, the method utilizes dipole array acoustic logging curves recorded before and after fracturing. Manual labeling was conducted by integrating logging data interpretation results. An improved WGAN-GP was employed to generate adversarial samples for data augmentation, and fracturing effect evaluation was implemented using SE-ResNet, ResNet, and DenseNet. The experimental results demonstrated that ResNet with residual connections is more suitable for the dataset in this study, achieving higher accuracy in fracturing effect evaluation. The inclusion of the SE module further enhanced model accuracy by adaptively adjusting the weights of feature map channels, with the highest accuracy reaching 99.75%.
... Machine learning is mainly utilized in this domain for the evaluation of reservoir stimulation, as well as production enhancement. For example, Erofeev et al. [31] investigated the applicability of the gradient boosting machine learning algorithm in predicting oil and total liquid production after hydraulic fracturing by examining the production data from more than 2000 fractured wells. This method can also be used as a new approach for HF candidate selection based on real-time field performance. ...
With the rapid progress of big data and artificial intelligence, machine learning technologies such as learning and adaptive control have emerged as a research focus in petroleum engineering. They have various applications in oilfield development, such as parameter prediction, optimization scheme deployment, and performance evaluation. This paper provides a comprehensive review of these applications in three key scenarios of petroleum engineering, namely hydraulic fracturing and acidizing, chemical flooding and gas flooding, and water injection. This article first introduces the steps and methods of machine learning processing in these scenarios, then discusses the advantages, disadvantages, existing challenges, and future prospects of these machine learning methods. Furthermore, this article compares and contrasts the strengths and weaknesses of these machine learning methods, aiming to help researchers select and improve their methods. Finally, this paper identifies some potential development trends and research directions of machine learning in petroleum engineering based on the current issues.
... The Oil and Gas industry has been undergoing a recent digitalization process [13]. In this context, AI techniques have been widely used in several sectors of the industry [4,5,24]. One of the areas where there is great potential for the application of AI techniques, especially ML, is in the decommissioning of wells. ...
Due to the growth of Plugging and Abandonment operations, the challenges of assessing the integrity of the cement layer and the quality of its bond to the casing and formation increase consequentially. Hence, it is paramount to ensure that the wellbore is hydraulically isolated from the surrounding environment before permanently sealing the well. However, nowadays, this process depends on the skills of a specialist interpreting a vast amount of complex data acquired through logging operations, which turns the task human-dependent, error-prone, and time-consuming. Motivated by that cement evaluation task, ouronova, in partnership with Repsol Sinopec Brazil, is developing a computational tool to interactively assist the specialist in interpreting cement integrity logging data and the operator in optimizing the planning and management of Plugging and Abandonment campaigns. The so-called P&A Assistant software uses machine learning techniques that, through the work done so far, have shown to be a promising alternative to improve the accuracy and reliability and reduce the time of the cement sheath integrity analysis. The software is also prepared to work with logging data acquired in a through-tubing configuration, which represents a reduction in operational cost and time. The paper presents the software's initial module, presenting three different unsupervised methods (K-means, Bisecting K-means, and Gaussian Mixture Model) and input feature combinations, with the aim of optimizing the model. The main results of the work indicate that the methods implemented using the Cement Bond Long channel and Bond Index channel have better results when compared to the models combined with Variable Density Log and AIBK, with values above 0.7 for Rand Index and 0.5 for Silhouette Coefficient. For the unsupervised methods, the K-mean model had the best performance.
... Recent developments in the technology of artificial intelligence have offered new views to revisit many traditional but important problems in the petroleum industry (Bravo, 2014;Amr et al., 2018;Rahmanifard and Plaksina, 2019;Erofeev et al., 2021;Wang et al., 2021a). For the prediction of remaining oil, Masini et al. (2019) presented an innovative workflow combing the classical reservoir engineering and the Locate-the-Remaining-Oil (LTRO) techniques with the smart data science and artificial neural network. ...
The remaining oil distribution plays an important role in enhanced oil recovery (EOR), which directly guides the development of an oil reservoir in the middle and later stages. However, it is still challenging to accurately and efficiently characterize the distribution of remaining oil due to the complex reservoir geology. We propose a data-driven physics-informed interpolation evolution algorithm combining historical-predicted knowledge (DPIE-HPK) for the prediction of remaining oil distribution totally based on reservoir monitoring data. As a key step towards the remaining oil distribution, the production rates can be predicted as well. A physics-informed data supplement (PIDS) process is also presented to assist the DPIE-HPK algorithm. Both historical physics information and future information are used in the DPIE-HPK framework. The historical physical information is preprocessed by the PIDS, and future information is predicted by Long Short-Term Memory (LSTM) deep learning models. As a crucial interpolation evolution method, the Kriging method is also integrated into the framework to evolve the unknown spatial information. We test the DPIE-HPK framework for the prediction of the remaining oil distribution of a typical tight carbonate oil reservoir located in the Middle East. It is concluded that the DPIE-HPK framework is a fast, accurate and efficient intelligence tool for predicting the remaining oil distribution.
... The feature-based machine learning methods such as linear regression require manual feature engineering, leading to the loss of important information contained in raw data. In the literature, different authors also applied the widespread machine learning techniques such as the Gradient Boosting algorithm and/or Artificial Neural Network to predict the bottomhole pressure [9,10] of a well and the flow rate after the hydraulic fracturing treatment [11,12,13]. ...
We propose a novel approach to data-driven modeling of a transient production of oil wells. We apply the transformer-based neural networks trained on the multivariate time series composed of various parameters of oil wells measured during their exploitation. By tuning the machine learning models for a single well (ignoring the effect of neighboring wells) on the open-source field datasets, we demonstrate that transformer outperforms recurrent neural networks with LSTM/GRU cells in the forecasting of the bottomhole pressure dynamics. We apply the transfer learning procedure to the transformer-based surrogate model, which includes the initial training on the dataset from a certain well and additional tuning of the model's weights on the dataset from a target well. Transfer learning approach helps to improve the prediction capability of the model. Next, we generalize the single-well model based on the transformer architecture for multiple wells to simulate complex transient oilfield-level patterns. In other words, we create the global model which deals with the dataset, comprised of the production history from multiple wells, and allows for capturing the well interference resulting in more accurate prediction of the bottomhole pressure or flow rate evolutions for each well under consideration. The developed instruments for a single-well and oilfield-scale modelling can be used to optimize the production process by selecting the operating regime and submersible equipment to increase the hydrocarbon recovery. In addition, the models can be helpful to perform well-testing avoiding costly shut-in operations.
... The feature-based machine learning methods such as linear regression require manual feature engineering, leading to the loss of important information contained in raw data. In the literature, different authors also applied the widespread machine learning techniques such as the Gradient Boosting algorithm and/or Artificial Neural Network to predict the bottomhole pressure [9,10] of a well and the flow rate after the hydraulic fracturing treatment [11,12,13]. ...
We propose a novel approach to data-driven modeling of a transient production of oil wells. We apply the transformer-based neural networks trained on the multivariate time series composed of various parameters of oil wells measured during their exploitation. By tuning the machine learning models for a single well (ignoring the effect of neighboring wells) on the open-source field datasets, we demonstrate that transformer outperforms recurrent neural networks with LSTM/GRU cells in the forecasting of the bottomhole pressure dynamics. We apply the transfer learning procedure to the transformer-based surrogate model, which includes the initial training on the dataset from a certain well and additional tuning of the model's weights on the dataset from a target well. Transfer learning approach helps to improve the prediction capability of the model. Next, we generalize the single-well model based on the transformer architecture for multiple wells to simulate complex transient oilfield-level patterns. In other words, we create the global model which deals with the dataset, comprised of the production history from multiple wells, and allows for capturing the well interference resulting in more accurate prediction of the bottomhole pressure or flow rate evolutions for each well under consideration. The developed instruments for a single-well and oilfield-scale modelling can be used to optimize the production process by selecting the operating regime and submersible equipment to increase the hydrocarbon recovery. In addition, the models can be helpful to perform well-testing avoiding costly shut-in operations.
This paper investigates the automation of hydraulic fracturing within the context of Industry 4.0, defining intelligent fracturing systems through three components: surface equipment, downhole sensing, and software control systems. The study addresses major advancements such as real-time optimization of fracturing parameters, intelligent allocation of fracturing fluids and materials, risk early warning systems, and fracture monitoring. It identifies four technical challenges: integrating heterogeneous data, developing intelligent decision-making algorithms, adaptive surface equipment adjustments, and multi-machine collaborative control. The paper reviews technological progress, foresees future benchmarks, and predicts the potential for unmanned fracturing operations, including site reconnaissance, autonomous fracturing unit driving, intelligent wellsite layout, fluid mixing, and operation execution. This comprehensive review delineates the developmental framework of intelligent fracturing, outlines key challenges, and forecasts future trends, marking a significant contribution to the field.
In this comprehensive study, machine learning (ML) techniques are employed to revolutionize lithology classification within the geosciences, emphasizing the transformative impact of ML on traditional practices. The research encapsulates ML's integration into well-log data analysis, enhancing prediction accuracy and efficiency in lithology identification—a crucial aspect of subsurface exploration.
The methodology adopted includes systematic data preprocessing, feature extraction, and the deployment of advanced ML algorithms such as Support Vector Machines and Random Forest for lithology classification. Models are trained and validated against well-log data from the Teapot Dome Reservoir and the Force 2020 Dataset, with the latter representing a collaborative and competitive environment aimed at advancing ML applications in geoscience.
Results reveal a marked increase in predictive accuracy when incorporating a wider array of logs, as evidenced by Models A1 and A2 for the Teapot Dome Reservoir, and Models B1 and B2 for the Force 2020 Dataset. The research highlights the critical role of ML in achieving high accuracies in lithology prediction, with improved generalization capabilities across different geological settings.
The workflow emphasizes the potential of ML algorithms to enhance well-log interpretation, streamline geological analyses, and reduce the time required for data processing. The study suggests future work focusing on expanding lithology types, normalizing log data, and broadening geographical coverage to further refine ML models for lithology classification. This effort underscores the convergence of ML with geoscience, promising a future where digital technologies create a more interconnected system for subsurface exploration.
Drilling rigs are an expensive resource in the oil and gas industry; hence, planning their time properly is necessary. Estimating activity durations plays a crucial role in the rigs' planning. The selection of correlated wells to be used in estimation models is vital for good duration estimative. Based on this necessity, we present a regression supervised method to cluster wells. Results were compared with traditional unsupervised clustering methods.
The developed method transforms a machine learning unsupervised problem (without a target) into a machine learning regression supervised problem (with a target). We use a decision tree regression to transform an unsupervised clustering problem into a supervised clustering problem and compare it with the other two methods based on K-means (unsupervised). We apply multiple linear regression with leave-one-out cross-validation in the resulting clusters to evaluate the clusters.
We tested the methodology in eight different scenarios, varying in number of wells (42 to 236) and similarities. All scenarios were based on real wells from a large Brazilian energy company. The target for the decision tree regression methodology was the drilling duration, and the variables were water depth and well footage. We limited the minimum number of wells per cluster, reducing overfitting in the estimation technique. Our methodology reached the lowest mean absolute percentage error value for all scenarios. The decision tree regression methodology can increase the capability of the expert to select similar wells based on their characteristics and duration.
The novelty of the decision tree regression method applied in a clustering problem is to provide the experts with a tool that considers the operational duration, besides the traditional set of relevant variables, transforming an unsupervised problem into a regression supervised problem. With the decision tree regression method, it is possible to cluster wells in more similar groups, improving any further analysis.
The reliable and predictive control for closed-loop refracturing design and optimization is of significance to improve the unconventional resource development and recovery in petroleum engineering. To reduce refracturing-decision risk under uncertainty for unconventional shale gas reservoir, we present an efficient and robust two-stage refracturing optimization workflow through hybridizing data-space inversion and non-dominated sorting genetic algorithm-II. Without performing history matching step, we adapt data-space inversion to simultaneously predict post-history pressure field and shale gas production corresponding to specific refracturing parameters given history measurement. In the proposed framework, both the enhanced net present value and enhanced cumulative gas production after refracturing stimulation are regarded as the bi-objective functions while the refracturing parameters including fracture number and fracture half-length are chosen as the decision variables. The proposed two-stage refracturing optimization strategy enables us to effectively identify the potential candidate clusters and/or wells and then search the optimal refracturing parameters efficiently and reasonably.
We disclose a new-age field-scale production forecast model that handles complex treatment of wellbores during their life cycle. Predictive production models have been an object of increased interest and research for a long time due to the need for a fast tool for forecasting production rates or choosing an optimal field development scheme. The existing approaches based on the material balance equation have several limitations and are not very applicable for real objects. Full-scale reservoir modeling is relatively slow and requires large computing resources. In this paper, we propose a proxy model based on advanced capacitance-resistance approach. The model predicts multiphase flow rates based on the available historical data of field production and information about well treatments. In addition, it provides preferable transmissibility trends, the presence of sealed or leaking faults, and the degree of dissipation between injector-producer well pairs. The advanced feature of the model is time-dependent weight coefficients, which have not been studied previously. They help in accounting the shut-in and workover periods and can be found during the optimization procedure simultaneously. Another feature is fast calculations due to a vectorized form of the model and application of modern optimization techniques. All these options allow modeling real oil fields with a large number of wells and a complex system of production control.
Coalbed methane (CBM) has emerged as one of the clean unconventional resources to supplement the rising demand of oil and gas. Analyzing and predicting CBM production performance are critical in choosing the optimal completion methods and parameters. However, the conventional numerical simulation has challenges of complicated gridding issues and expensive computational costs. The huge amount of available production data that has been collected in the field site opens up a new opportunity to develop data-driven approaches in predicting the production rate. Here, we proposed a novel physics-constrained data-driven workflow to effectively forecast the CBM productivity based on a gated recurrent unit (GRU) and multilayer perceptron (MLP) combined neural network (GRU-MLP model). The model architecture is optimized automatically by the multiobjective algorithm: nondominated sorting genetic algorithm Ⅱ (NSGA Ⅱ). The proposed framework was used to predict gas and water production in synthetic cases with various fracture-network-complexity/connectivity and two multistage fractured horizontal wells in field sites located at Ordos Basin and Qinshui Basin, China. The results indicated that the proposed GRU-MLP combined neural network was able to accurately and stably predict the production performance of CBM fractured wells in a fast manner. Compared with recurrent neural network (RNN), GRU, and long short-term memory (LSTM), the proposed GRU-MLP had the highest accuracy, stability, and generalization, especially in the peak or trough and late-time production periods, because it could capture the production-variation trends precisely under the static and dynamic physical constraints. Consequently, a physics-constrained data-driven approach performed better than a pure data-driven method. Moreover, the contributions of constraints affecting the model prediction performance were clarified, which could provide insights for the practicing engineers to choose which categorical constraints are needed to focus on and preferentially treated if there are uncertainties and unknowns in a realistic reservoir. In addition, the optimum GRU-MLP model architecture was a group of optimized solutions, rather than a single solution. Engineers can evaluate the tradeoffs within this optimal set according to the field-site requirements. This study provides a novel machine learning approach based on a GRU-MLP combined neural network to estimate production performances in naturally fractured reservoir. The method is gridless and simple, but is capable of predicting the productivity in a computational cost-effective way. The key findings of this work are expected to provide a theoretical guidance for the intelligent development in oil and gas industry.
Coalbed methane (CBM) has emerged as one of the clean unconventional resources to supplement the rising demand of conventional hydrocarbons. Analyzing and predicting CBM production performance is critical in choosing the optimal completion methods and parameters. However, the conventional numerical simulation has challenges of complicated gridding issues and expensive computational costs. The huge amount of available production data that has been collected in the field site opens up a new opportunity to develop data-driven approaches in predicting the production rate. Here, we proposed a novel physics-constrained data-driven workflow to effectively forecast the CBM productivity based on a Gated Recurrent Unit (GRU) and Multi-Layer Perceptron (MLP) combined neural network (GRU-MLP model). The model architecture is optimized by the multiobjective algorithm: nondominated sorting genetic algorithm Ⅱ (NSGA Ⅱ). The proposed framework was used to predict synthetic cases with various fracture-network-complexities and two multistage-fractured wells in field sites located at Qinshui basin and Ordos basin, China. The results indicated that the proposed GRU-MLP combined neural network was able to accurately and stably predict the production performance of multi-fractured horizontal CBM wells in a fast manner. Compared with Simple Recurrent Neural Network (RNN), Gated Recurrent Unit (GRU) and Long Short-Term Memory (LSTM), the proposed GRU-MLP had the highest accuracy and stability especially for gas production in late-time. Consequently, a physics-constrained data-driven approach performed better than a pure data-driven method. Moreover, the optimum GRU-MLP model architecture was a group of optimized solutions, rather than a single solution. Engineers can evaluate the tradeoffs within this set according to the field-site requirements. This study provides a novel machine learning approach based on a GRU-MLP combined neural network model to estimate production performances in CBM wells. The method is simple and gridless, but is capable of predicting the productivity in a computational cost-effective way. The key findings of this work are expected to provide a theoretical guidance for the intelligent development in oil and gas industry.
We describe a stacked model for predicting the cumulative fluid production for an oil well with a multistage-fracture completion based on a combination of Ridge Regression and CatBoost algorithms. The model is developed based on an extended digital field data base of reservoir, well and fracturing design parameters. The database now includes more than 5000 wells from 23 oilfields of Western Siberia (Russia), with 6687 fracturing operations in total. Starting with 387 parameters characterizing each well, including construction, reservoir properties, fracturing design features and production, we end up with 38 key parameters used as input features for each well in the model training process. The model demonstrates physically explainable dependencies plots of the target on the design parameters (number of stages, proppant mass, average and final proppant concentrations and fluid rate). We developed a set of methods including those based on the use of Euclidean distance and clustering techniques to perform similar (offset) wells search, which is useful for a field engineer to analyze earlier fracturing treatments on similar wells. These approaches are also adapted for obtaining the optimization parameters boundaries for the particular pilot well, as part of the field testing campaign of the methodology. An inverse problem (selecting an optimum set of fracturing design parameters to maximize production) is formulated as optimizing a high dimensional black box approximation function constrained by boundaries and solved with four different optimization methods: surrogate-based optimization, sequential least squares programming, particle swarm optimization and differential evolution. A recommendation system containing all the above methods is designed to advise a production stimulation engineer on an optimized fracturing design.
We analyse the influence of fluid yield stress on the propagation of a radial (penny-shaped) hydraulic fracture in a permeable reservoir. In particular, the Herschel-Bulkley rheological model is adopted that includes yield stress and non-linearity of the shear stress. The rock is assumed to be linear elastic, and the fracture is driven by the point source fluid injection with a constant volumetric rate. The fracture propagation condition follows the theory of linear elastic fracture mechanics, and Carter's leak-off law is selected to govern the fluid exchange process between the fracture and formation. The numerical solution of the problem is found using the algorithm based on Gauss-Chebyshev quadrature and Barycentric Lagrange interpolation techniques. We also construct an approximate solution with the help of the global fluid balance equation and the near-tip region asymptote. The latter approximation is computationally efficient, and we estimate its accuracy by comparing the primary crack characteristics such as opening, pressure, and radius with those provided by the full numerical solution. We present examples corresponding to typical field cases and demonstrate that the addition of yield stress can lead to a shorter radius and wider opening compared to the corresponding case with simpler power-law fluid rheology. Further, we quantify the limiting propagation regimes (or vertex solutions) characterised by the dominance of a particular physical phenomenon. Relative to the power-law results, there are two new vertices that are associated with the domination of yield stress: storage-yield-stress and leak-off-yield-stress. To understand the influence of various problem parameters, we utilise the constructed approximate solution to investigate the dimensionless parametric space of the problem, in which the applicability domains of the limiting solutions are quantified. This enables one to quickly determine whether yield stress provides a strong influence for given problem parameters.
The quasi-brittle nature of rocks challenges the basic assumptions of linear hydraulic fracture mechanics (LHFM): namely, linear elastic fracture mechanics and smooth parallel plates lubrication fluid flow inside the propagating fracture. We relax these hypotheses and investigate in details the growth of a plane-strain hydraulic fracture in an impermeable medium accounting for a rough cohesive zone and a fluid lag. In addition to a dimensionless toughness and the time-scale tom of coalescence of the fluid and fracture fronts governing the fracture evolution in the LHFM case, the solution now also depends on the ratio between the in-situ stress and material peak cohesive stress σo∕σc and the intensity of the flow deviation induced by aperture roughness (captured by a dimensionless power exponent). We show that the solution is appropriately described by a nucleation time-scale tcm=tom×(σo∕σc)3, which delineates the fracture growth into three distinct stages: a nucleation phase (t≪tcm), an intermediate stage (t∼tcm) and late time (t≫tcm) stage where convergence toward LHFM predictions finally occurs. A highly non-linear hydro-mechanical coupling takes place as the fluid front enters the rough cohesive zone which itself evolves during the nucleation and intermediate stages of growth. This coupling leads to significant additional viscous flow dissipation. As a result, the fracture evolution deviates from LHFM predictions with shorter fracture lengths, larger widths and net pressures. These deviations from LHFM ultimately decrease at late times (t≫tcm) as the ratios of the lag and cohesive zone sizes with the fracture length both become smaller. The deviations increase with larger dimensionless toughness and larger σo∕σc ratio, as both have the effect of further localizing viscous dissipation near the fluid front located in the small rough cohesive zone. The convergence toward LHFM can occur at very late time compared to the nucleation time-scale tcm (by a factor of hundred to thousand times) for realistic values of σo∕σc encountered at depth. The impact of a rough cohesive zone appears to be prominent for laboratory experiments and short in-situ injections in quasi-brittle rocks with ultimately a larger energy demand compared to LHFM predictions.
We analyze how artificial intelligence changes a significant part of the energy sector, the oil and gas industry. We focus on the upstream segment as the most capital-intensive part of oil and gas and the segment of enormous uncertainties to tackle. Basing on the analysis of AI application possibilities and the review of existing applications, we outline the most recent trends in developing AI-based tools and identify their effects on accelerating and de-risking processes in the industry. We investigate AI approaches and algorithms, as well as the role and availability of data in the segment. Further, we discuss the main non-technical challenges that prevent the intensive application of artificial intelligence in the oil and gas industry, related to data, people, and new forms of collaboration. We also outline three possible scenarios of how artificial intelligence will develop in the oil and gas industry and how it may change it in the future (in 5, 10, and 20 years).
The bidirectional long short-term memory (Bi-LSTM) network is an innovative computation paradigm that learns bidirectional long-term dependencies between time steps and sequence data to predict future occurrences. This study proposes a framework to incorporate Bi-LSTM and data sequencing to predict diameter of jet grouted columns in soft soil in real time. The models are tested using a case study of jet grouting treatment of soft soil. The results show that the proposed strategies can efficiently predict the variation in column diameter with the depth. A comparative performance analysis among the Bi-LSTM, original long short-term memory (LSTM) and support vector regression (SVR) approaches is also conducted. The Bi-LSTM performs better than both the LSTM and SVR in root-mean-square error. This result substantiates the efficacy of modeling sequential step-by-step jet grouting process using the Bi-LSTM. Based on the analyzed results, some recommendations for improving the current design of jet grout columns are proposed.
This paper investigates the problem of a radial (or penny-shaped) hydraulic fracture propagating in a permeable reservoir. In particular, we consider the fluid exchange between the crack and ambient porous media as a pressure-dependent process. In most of the existing models, the fluid exchange process is represented as one-dimensional pressure-independent leak-off described by Carter's law. We modify this mechanism by including the dependence of the fluid-exchange rate on the fluid pressure inside the fracture. The proposed approach allows the liquid not only to flow out of the fracture but also to leak-in along the region adjacent to the fracture front. The complete model for hydraulic fracturing involves several physical processes that determine the crack propagation, namely, brittle rock failure, elastic equilibrium of rock, viscous fluid flow inside the fracture channel, and fluid exchange. In order to resolve these phenomena near the fracture front in the numerical scheme, we utilise a special asymptotic multi-scale model for the near-tip region, which is used as a propagation condition for the finite fracture. The main aspect of the analysis is a comparison of fracture characteristics such as aperture profile, radius and others calculated via the proposed model with the results of the radial fracture model that assumes standard pressure-independent fluid exchange mechanism governed by Carter's law. Based on the comparison, we determine parameter ranges, for which the effect of the pressure-dependent fluid exchange mechanism is essential, and, on the other hand, outline zones for which Carter's leak-off model provides accurate results. Finally, by using an approximate solution for the radial fracture with Carter's leak-off, we perform estimates of the effect for the entire parametric space, which allows one to evaluate the influence of the pressure-dependent leak-off for any problem parameters.
Knowledge of permeability, a measure of the ability of rocks to allow fluids to flow through them, is essential for building accurate models of oil and gas reservoirs. Permeability is best measured in the laboratory using special core analysis (SCAL), but this is expensive and time-consuming. This is the first major work on predicting permeability in the in the UK Continental Shelf (UKCS) based only on routine core analysis (RCA) data and a machine-learning approach. We present a comparative analysis of the various machine learning algorithms and validate the results, using permeability measured on 273 core samples from 104 wells. Results suggest that machine learning can predict permeability with relatively high accuracy. This opens new research directions in particular in the oil and gas sector.
Rock thermal conductivity is an essential input parameter for enhanced oil recovery methods design and optimization and for basin and petroleum system modelling. Absence of any effective technique for direct in situ measurements of rock thermal conductivity makes the development of well-log based methods for rock thermal conductivity determination highly desirable. A major part of the existing problem solutions is regression model-based approaches. Literature review revealed that there are only several studies performed to assess the applicability of neural network-based algorithms to predict rock thermal conductivity from well-logging data. In this research, we aim to define the most effective machine-learning algorithms for well-log based determination of rock thermal conductivity. Well-logging data acquired at a heavy oil reservoir together with results of thermal logging on cores extracted from two wells were the basis for our research. Eight different regression models were developed and tested to predict vertical variations of rock conductivity from well-logging data. Additionally, rock thermal conductivity was determined based on Lichtenecker-Asaad model. Comparison study of regression-based and theoretical-based approaches was performed. Among considered machine learning techniques Random Forest algorithm was found to be the most accurate at well-log based determination of rock thermal conductivity. From a comparison of the thermal conductivity - depth profile predicted from well-logging data with the experimental data, and it can be concluded that thermal conductivity can be determined with a total relative error of 12.54%. The obtained results prove that rock thermal conductivity can be inferred from well-logging data for wells that are drilled in a similar geological setting based on the Random Forest algorithm with an accuracy sufficient for industrial needs.
The present work considers the development of a machine learning model trained on synthetic and lab data for the steady-state multiphase pipe flow. We propose a new method for calculating flow characteristics such as liquid holdup, flow regime, and pressure gradient in the pipe segment based on ANNs and transfer learning technique. Besides, the created tool is implemented within the marching algorithm for calculating flow parameters along the whole pipe. The segment module consists of three sub-models, namely, for calculating liquid holdup, defining flow regime, and estimating pressure gradient. For sub-models creation, we use transfer learning methodology: on the first stage, the ANNs are trained on synthetic data, which we generate by using OLGAS mechanistic model; on the second stage, we train meta-models additionally on the real data, which in our case presented by lab measurements. As a result, we create the new multiphase flow correlation, which includes the basics of the physics-based OLGAS model and is tuned for real data that can be in the general case field measurements. At the final stage, we apply marching algorithms with the suggested segment model to the field dataset for testing purposes.
In this paper we consider the near-tip region of a fluid-driven fracture propagating in permeable rock. We attempt to accurately resolve the coupling between the physical processes-rock breakage, fluid pressure drop in the viscous fluid flow in the fracture and fluid exchange between the fracture and the rock-that exert influence on the hydraulic fracture propagation, yet occur over length scales often too small to be efficiently captured in existing coarse grid numerical models. We consider three fluid balance mechanisms: storage in the fracture, pore fluid leak-in from the rock into the fracture as the result of dynamic suction at the dilating crack tip, and fluid leak-off from the fracture into the rock as the fluid pressure in the fracture recovers with distance away from the tip. This process leads to the formation of a pore fluid circulation cell adjacent to the propagating fracture tip. We obtain the general numerical solution for the fracture opening and fluid pressure in the semi-infinite steadily propagating fracture model, while assuming that the hydraulic fracturing and pore fluids have the same properties. We fully characterize the solution within the problem parametric space and identify different regimes of the fracture propagation. We assess the impact of the pore fluid leak-in and the associated near-tip circulation cavity on the solution and explore limitations of the widely used, pressure-independent Carter's leak-off model. The obtained solution could be potentially used as a tip element in the finite crack models (penny-shaped, Planar3D), provided that a fast numerical implementation is further elaborated.
Flow, transport, mechanical, and fracture properties of porous media depend on their morphology and are usually estimated by experimental and/or computational methods. The precision of the computational approaches depends on the accuracy of the model that represents the morphology. If high accuracy is required, the computations and even experiments can be quite time-consuming. At the same time, linking the morphology directly to the permeability, as well as other important flow and transport properties, has been a long-standing problem. In this paper, we develop a new network that utilizes a deep learning (DL) algorithm to link the morphology of porous media to their permeability. The network is neither a purely traditional artificial neural network (ANN), nor is it a purely DL algorithm, but, rather, it is a hybrid of both. The input data include three-dimensional images of sandstones, hundreds of their stochastic realizations generated by a reconstruction method, and synthetic unconsolidated porous media produced by a Boolean method. To develop the network, we first extract important features of the images using a DL algorithm and then feed them to an ANN to estimate the permeabilities. We demonstrate that the network is successfully trained, such that it can develop accurate correlations between the morphology of porous media and their effective permeability. The high accuracy of the network is demonstrated by its predictions for the permeability of a variety of porous media.
The description of rocks is one of the most time-consuming tasks in the everyday work of a geologist, especially when very accurate description is required. We here present a method that reduces the time needed for accurate description of rocks, enabling the geologist to work more efficiently. We describe the application of methods based on color distribution analysis and feature extraction. Then we focus on a new approach, used by us, which is based on convolutional neural networks. We used several well-known neural network architectures (AlexNet, VGG, GoogLeNet, ResNet) and made a comparison of their performance. The precision of the algorithms is up to 95% on the validation set with GoogLeNet architecture. The best of the proposed algorithms can describe 50 m of full-size core in one minute.
This work is devoted to an assessment of the application of machine learning algorithms in the prediction of a fracture's aspect ratio caused by the hydraulic fracturing. By the aspect ratio in this work is assumed the ratio of the larger half-axis of the fracture to the smaller one. The study shows the prospects of applying data-driven surrogate model methods (deep neural networks learning from data simulated by means of traditional solvers) to particle dynamics modelling of hydraulic fracturing. The solution obtained allows to predict the aspect ratio value quickly, and thus, to evaluate the volume of the hydraulic fracturing fluid injection necessary to achieve the required fracture length.
Engineering simulators used for steady-state multiphase flows in oil and gas wells and pipelines are commonly utilized to predict pressure drop and phase velocities. Such simulators are typically based on either empirical correlations (e.g., Beggs and Brill, Mukherjee and Brill, Duns and Ros) or first-principles mechanistic models (e.g., Ansari, Xiao, TUFFP Unified, Leda Flow Point model, OLGAS). The simulators allow one to evaluate the pressure drop in a multiphase pipe flow with acceptable accuracy. However, the only shortcoming of these correlations and mechanistic models is their applicability (besides steady-state versions of transient simulators such as Leda Flow and OLGA). Empirical correlations are commonly applicable in their respective ranges of data fitting; and mechanistic models are limited by the applicability of the empirically based closure relations that are a part of such models. In order to extend the applicability and the accuracy of the existing accessible methods, a method of pressure drop calculation in the pipeline is proposed. The method is based on well segmentation and calculation of the pressure gradient in each segment using three surrogate models based on Machine Learning (ML) algorithms trained on a representative lab data set from the open literature. The first model predicts the value of a liquid holdup in the segment, the second one determines the flow pattern, and the third one is used to estimate the pressure gradient. To build these models, several ML algorithms are trained such as Random Forest, Gradient Boosting Decision Trees, Support Vector Machine, and Artificial Neural Network, and their predictive abilities are cross-compared. The proposed method for pressure gradient calculation yields R 2 = 0.95 by using the Gradient Boosting algorithm as compared with R 2 = 0.92 in case of Mukherjee and Brill correlation and R 2 = 0.91 when a combination of Ansari and Xiao mechanistic models is utilized. The application of the above-mentioned ML algorithms and the larger database used for their training will allow extending the proposed methodology to a wider applicability range of input parameters as compared to standard accessible techniques. The method for pressure drop prediction based on ML algorithms trained on lab data is also validated on three real field cases. Validation indicates that the proposed model yields the following coefficients of determination: R 2 = 0.806, 0.815 and 0.99 as compared with the highest values obtained by commonly used techniques: R 2 = 0.82 (Beggs and Brill correlation), R 2 = 0.823 (Mukherjee and Brill correlation) and R 2 = 0.98 (Beggs and Brill correlation). Hence, the method for calculating the pressure distribution could give comparable or even higher scores on field data by contrast to correlations and mechanistic models. This fact is an indicator that the model can be scalable from the lab to the field conditions without any additional retraining of ML algorithms.
The objective of this work is to study the applicability of various machine learning algorithms for the prediction of some rock properties which geoscientists usually define due to special laboratory analysis. We demonstrate that these special properties can be predicted only basing on routine core analysis (RCA) data. To validate the approach, core samples from the reservoir with soluble rock matrix components (salts) were tested within 100 + laboratory experiments. The challenge of the experiments was to characterize the rate of salts in cores and alteration of porosity and permeability after reservoir desalination due to drilling mud or water injection. For these three measured characteristics, we developed the relevant predictive models, which were based on the results of RCA and data on coring depth and top and bottom depths of productive horizons. To select the most accurate machine learning algorithm, a comparative analysis has been performed. It was shown that different algorithms work better in different models. However, two-hidden-layer neural network has demonstrated the best predictive ability and generalizability for all three rock characteristics jointly. The other algorithms, such as support vector machine and linear regression, also worked well on the dataset, but in particular cases. Overall, the applied approach allows predicting the alteration of porosity and permeability during desalination in porous rocks and also evaluating salt concentration without direct measurements in a laboratory. This work also shows that developed approaches could be applied for the prediction of other rock properties (residual brine and oil saturations, relative permeability, capillary pressure, and others), of which laboratory measurements are time-consuming and expensive.
In this paper, we present a data-driven model for forecasting the production increase after hydraulic fracturing (HF). We use data from fracturing jobs performed at one of the Siberian oilfields. The data includes features, characterizing the jobs, and geological information. To predict an oil rate after the fracturing, machine learning (ML) technique was applied. We have compared the ML-based prediction to a prediction based on the experience of reservoir and production engineers responsible for the HF-job planning. We discuss the potential for further development of ML techniques for predicting changes in oil rate after HF.
We present a research study aimed at testing of applicability of machine learning techniques for prediction of permeability of digitized rock samples. We prepare a training set containing 3D images of sandstone samples imaged with X-ray microtomography and corresponding permeability values simulated with Pore Network approach. We also use Minkowski functionals and Deep Learning-based descriptors of 3D images and 2D slices as input features for predictive model training and prediction. We compare predictive power of various feature sets and methods. The later include Gradient Boosting and various architectures of Deep Neural Networks (DNN). The results demonstrate applicability of machine learning for image-based permeability prediction and open a new area of Digital Rock research.
Engineering tools allowing pressure gradient calculation in the pipe segment commonly use stationary correlation and mechanistic models such as Beggs and Brill, Ansary, etc. This is well known and convenient way which gives rough estimate of pressure gradient due to friction losses and liquid phase interference. It avoids solving complex dynamic pressure equation and derives quick results with comfortable precision margin for large scale systems, such as horizontal pipes and wells. In order to enlarge the applicability zone and accuracy of existing methods, a new method of pressure gradient definition is evolved. It is included three surrogate models that are based on Machine Learning (ML) algorithms. The first model predicts liquid holdup in the segment, the second defines flow pattern and the third predicts pressure gradient. In order to create these models, several ML algorithms are applied such as Random Forest, Gradient Boosting, Support Vector Machine and Artificial Neuron Network.
Involvement of the latest machine learning algorithms will allow applying this method to wider range of input data compared with standard multiphase flow correlations and mechanistic models. The proposed method demonstrates high accuracy – on the collected experimental data set it gives R ² = 0.985 for pressure gradient prediction. That is why it could help to carry out correct calculation of bottom hole pressure and pressure distribution along the length of the pipeline.
This paper deals with the comparison of models for predicting porosity and permeability of oil reservoirs by coupling a machine learning concept and petrophysical logs. Different machine learning methods including conventional artificial neural network, genetic algorithm, fuzzy decision tree, the imperialist competitive algorithm (ICA), particle swarm optimization (PSO), and a hybrid of those ones are employed to have a comprehensive comparison. The machine learning approach was constructed and tested via data samples recorded from northern Persian Gulf oil reservoirs. The results gained from the machine learning models used in this paper are compared to the relevant real petrophysical data and the outputs achieved by other methods employed in our previous studies. The average relative absolute deviation between the approach estimations and the relevant actual data is found to be less than 1% for the hybridized approaches. The results reported in this paper indicate that implication of hybridized machine learning methods in porosity and permeability estimations can lead to the construction of more reliable static reservoir models in simulation plans.
We introduce the hyperparameter search problem in the field of machine
learning and discuss its main challenges from an optimization perspective.
Machine learning methods attempt to build models that capture some element of
interest based on given data. Most common learning algorithms feature a set of
hyperparameters that must be determined before training commences. The choice
of hyperparameters can significantly affect the resulting model's performance,
but determining good values can be complex; hence a disciplined, theoretically
sound search strategy is essential.
Function approximation is viewed from the perspective of numerical optimization in function space, rather than parameter space. A connection is made between stagewise additive expansions and steepest--descent minimization. A general gradient--descent "boosting" paradigm is developed for additive expansions based on any fitting criterion. Specific algorithms are presented for least--squares, least--absolute--deviation, and Huber--M loss functions for regression, and multi--class logistic likelihood for classification. Special enhancements are derived for the particular case where the individual additive components are decision trees, and tools for interpreting such "TreeBoost" models are presented. Gradient boosting of decision trees produces competitive, highly robust, interpretable procedures for regression and classification, especially appropriate for mining less than clean data. Connections between this approach and the boosting methods of Freund and Shapire 1996, and Fr...
Sensitivity to derivatives and a need for proper initial guesses are the main disadvantages of classic nonlinear solvers like Newton’s method. To overcome the obstacles, a numerical solver for second-order nonlinear Partial Differential Equations (PDEs) based on an Adaptive Neural Network (AdNN) is introduced. While using Newton’s method needs to form the Jacobian matrix and its inversion, which both of them are time-consuming and dependent on the nature of the problem, the proposed solver tries to find the roots of the algebraic equations by changing the weights of AdNN by the help adaptive laws. The proposed approach has been applied to solve the governing PDEs of the gas flow in shale resources and the immiscible two-phase flow of water and oil in hydrocarbon reservoirs as two highly nonlinear phenomena. The generated profiles of pressures and saturations show a satisfying match with the outputs of Newton’s method. However, using the presented algorithm not only removes the former necessities but also helps the community to solve the relevant PDEs governing the critical elements of the energy market in the future with a higher level of confidence.
This paper presents a new trial to reproduce soil stress–strain behaviour by adapting a long short-term memory (LSTM) deep learning method. LSTM is an approach that employs time sequence data to predict future occurrences, and it can be used to consider the stress history of soil behaviour. The proposed LSTM method includes the following three steps: data preparation, architecture determination, and optimisation. The capacity of the adapted LSTM method is compared with that of feedforward and feedback neural networks using a new numerical benchmark dataset. The performance of the proposed LSTM method is verified through a dataset collected from laboratory tests. The results indicate that the LSTM deep-learning method outperforms the feed forward and feedback neural networks based on both accuracy and the convergence rate when reproducing the soil’s stress–strain behaviour. One new phenomenon referred to as “bias at low stress levels”, which was not noticed before, is first discovered and discussed for all neural network-based methods.
This study investigates the problem of a hydraulic fracture propagating in a permeable formation with an emphasis on the computation of the poroelastic stresses caused by fluid leak-off into the formation. By using the assumption of smallness of the diffusion length scale relative to the fracture size, the three-dimensional poroelastic problem around the fracture splits into two simpler problems, one involving the undrained response due to fracture opening, and another one involving one-dimensional fluid flow in the direction perpendicular to the fracture and poroelastic stresses caused by an elevated pore pressure around the fracture. The latter leak-off induced stresses are often ignored due to difficulties associated with the complexity and/or computational demands needed to solve the problem. This study addresses this issue and presents an efficient solution for the problem that also makes it possible to better understand the influence of the leak-off induced stresses on hydraulic fracture propagation for large scale problems. It is shown that for the case of a planar fracture, the leak-off induced stress can be represented via an additional effective width, whose value is proportional to the total fluid loss. However, this does not apply for the general case of multiple hydraulic fractures since the off-plane spatial stress variation due to the pore pressure change and fracture opening are different (even though coincide on the fracture plane). To illustrate the developed concept, several numerical examples for various fracture configurations are presented.
Directional oil well drilling requires high precision of the wellbore positioning inside the productive area. However, due to specifics of engineering design, sensors that explicitly determine the type of the drilled rock are located farther than 15m from the drilling bit. As a result, the target area runaways can be detected only after this distance, which in turn, leads to a loss in well productivity and the risk of the need for an expensive re-boring operation.
We present a novel approach for identifying rock type at the drilling bit based on machine learning classification methods and data mining on sensors readings. We compare various machine-learning algorithms, examine extra features coming from mathematical modeling of drilling mechanics, and show that the real-time rock type classification error can be reduced from 13.5% to 9%. The approach is applicable for precise directional drilling in relatively thin target intervals of complex shapes and generalizes appropriately to new wells that are different from the ones used for training the machine learning model.
Elsevier author sharelink:
https://doi.org/10.1016/j.petrol.2020.107504
Growing amount of hydraulic fracturing (HF) jobs in the recent two decades resulted in a significant amount of measured data available for development of predictive models via machine learning (ML). In multistage fractured completions, post-fracturing production analysis (e.g., from production logging tools) reveals evidence that different stages produce very non-uniformly, and up to 30% may not be producing at all due to a combination of geomechanics and fracturing design factors. Hence, there is a significant room for improvement of current design practices. We propose a data-driven model for fracturing design optimization, where the workflow is essentially split into two stages. As a result of the first stage, the present paper summarizes the efforts into the creation of a digital database of field data from several thousands of multistage HF jobs on vertical, inclined and near-horizontal wells from circa 20 different oilfields in Western Siberia, Russia. In terms of the number of points (fracturing jobs), the present database is a rare case of a representative dataset of about 5000 data points, compared to typical databases available in the literature, comprising tens or hundreds of points at best. Each point in the data base contains the vector of 92 input variables (the reservoir, well and the frac design parameters) and the vector of production data, which is characterized by 16 parameters, including the target, cumulative oil production. The focus is made on data gathering from various sources, data preprocessing and development of the architecture of the database as well as solving the production forecast problem via ML. Data preparation has been done using various ML techniques: the problem of missing values in the database is solved with collaborative filtering for data imputation; outliers are removed using visualization of cluster data structure by t-SNE algorithm. The production forecast problem is solved via CatBoost algorithm. Prediction capability of the model is measured with the coefficient of determination (R2) and reached 0.815. The inverse problem (selecting an optimum set of fracturing design parameters to maximize production) will be considered in the second part of the study to be published in another paper, along with a recommendation system for advising DESC and production stimulation engineers on an optimized fracturing design.
This article proposes an approach to resolve the dynamic fracture of brittle materials by incorporating eigenerosion into the material point method (MPM) framework. The eigenerosion approach links the crack propagation to energy conservation based on the variational theory of fracture mechanics. This idea closely resembles the conventional treatment for the phase‐field method. The major difference is that the effective energy release rate of each particle that controls the crack propagation is only calculated within its neighborhood domain for the eigenerosion approach. Because evaluation of the material's fracture behavior can be decoupled from the governing equations as a separate solution step, the eigenerosion scheme allows straightforward implementation into any standard MPM solver with minor modifications. In addition, a phantom‐node method is employed to handle the preexisting crack. With these settings, the proposed model can capture complex fracture behaviors. Several representative benchmark tests demonstrate the efficiency and validity of the proposed model.
This paper considers the development of a computationally fast model for simulation of multiphase flow in porous media for a heterogeneous reservoir with the unlimited number of wells characterized by a different type of completion. This fast solution has been obtained by means of replacing the differential equation governing the flow in porous media by approximate governing equations which are parametrized by convolutional neural networks. The matching of the dynamic properties of the original and reduced models is ensured by conservation of spatial invariance property of the equations. The suggested approach is characterized by the minimal number of limitations and shortcomings related to geological-hydrodynamical structure and size of the original model. Also, there is no necessity of additional model training for reservoirs not included in a training dataset. Suggested approach has been evaluated on the synthetic benchmark test model SPE10, where a significant decrease in computational time has been demonstrated comparing to a traditional commercial reservoir simulator. Based on the results of all demonstrated test case scenarios, it could be noted that hybrid hydrodynamic modeling leads to a significant reduction in computational cost (by a factor of few hundreds), maintaining at the same time required accuracy of calculations.
Well-log depth matching is a long-standing challenge within the oil industry despite its importance in developing log interpretation algorithms exploiting correlations between different measurements. Gamma-ray logs are widely used as a proxy to match the depth of measurements acquired from different logging passes. Existing approaches are either manual or algorithm-assisted and are based on correlations. None performs well without user intervention. We have developed a supervised machine-learning-based solution that uses clever problem abstraction and data formation to alleviate the difficulty of the problem. A fully connected neural network was trained on data labeled through manual depth matching of field data with label-preserving data augmentation. A relaxed-accuracy criterion was adopted to improve the training effectiveness to deal with the unavoidable human error during manual labeling. Stacking techniques were employed to guarantee the robustness of this method. The solution led to well synchronized signals and made automation possible for the depth-matching process. The current development mainly focuses on wireline applications for vertical or low-deviation wells. © 2019 Society of Well Log Analystists Inc. All rights reserved.
With rapid development of unconventional tight and shale reservoirs, considerable amounts of data sets are increasing rapidly. Data mining techniques are becoming attractive alternatives for well performance evaluation and optimization. This paper develops a comprehensive data mining process to evaluate well production performance in Montney Formations in western Canadian sedimentary basin. The general data visualization and statistical data evaluation are used to qualitatively and quantitatively evaluate the relationships between the stimulation parameters and first-year oil production. Then, the recursive feature elimination with cross validation (RFECV) is used to identify the most important factors on the first-year oil production in unconventional reservoirs. In addition, four commonly used supervised learning approaches including random forest (RF), adaptive boosting (AdaBoost), support vector machine (SVM), and neural network (NN) are compared to estimate the first-year well production. The results show that 6 features are the most important variables for constructing an accurate prediction model: well latitude, longitude, well true vertical depth (TVD), proppant pumped per well, well lateral length, and fluid injected per well. Compared to other algorithms, RF has the best prediction performance for the first-year oil production. Furthermore, such data-driven models are found to be very useful for reservoir engineers when designing hydraulic fracture treatments in Montney tight reservoirs.
Tree boosting is a highly effective and widely used machine learning method. In this paper, we describe a scalable end-to-end tree boosting system called XGBoost, which is used widely by data scientists to achieve state-of-the-art results on many machine learning challenges. We propose a novel sparsity-aware algorithm for sparse data and weighted quantile sketch for approximate tree learning. More importantly, we provide insights on cache access patterns, data compression and sharding to build a scalable tree boosting system. By combining these insights, XGBoost scales beyond billions of examples using far fewer resources than existing systems.
The support-vector network is a new learning machine for two-group classification problems. The machine conceptually implements the following idea: input vectors are non-linearly mapped to a very high-dimension feature space. In this feature space a linear decision surface is constructed. Special properties of the decision surface ensures high generalization ability of the learning machine. The idea behind the support-vector network was previously implemented for the restricted case where the training data can be separated without errors. We here extend this result to non-separable training data. High generalization ability of support-vector networks utilizing polynomial input transformations is demonstrated. We also compare the performance of the support-vector network to various classical learning algorithms that all took part in a benchmark study of Optical Character Recognition.
There are a number of considerations to be made in the process of designing a fracturing stimulation treatment. The reservoir deliverability, well producing systems, fracture mechanics, fracturing fluid characteristics, proppant transport mechanism, operational constraints and economics should be considered and integrated in order to obtain the most cost-effective design and to maximize the benefit of a well stimulation treatment.
The main purpose of this work is to develop an analytical scheme which mathematically couples each elements as described above. The criteria used to determine the optimum size of the treatment are: (1) optimize reservoir deliverability, (2) maximize propped fracture penetration, (3) optimize pumping parameters, (4) minimize treatment cost, and (5) maximize economic returns of the well.
Firstly, an analytical inverse algorithm has been derived based on 2-D PKNC (Perkins, Kern, Nordgren and Carter) and KZGD (Khristianovic, Zheltov, Geertsma and de Klerk) formulations. It allows us to calculate the fluid and proppant volume needed for a desired fracture geometry and conductivity. Secondly, a technique to search for optimal pumping parameters and to maximize the proppant coverage for a given hydraulic penetration is developed. This allows us to optimize the propped geometry taking into account the operational constraints. Thirdly, a coupling algorithm is developed to link the reservoir producibility, well producing systems and the optimized fracture geometry. This enables us to optimize the wellhead deliverability based on the balance between the reservoir and the fracture characteristics. Finally, the overall economic analysis is performed to calculate the net present value for various design options. The most cost-effective treatment design can then be determined based on the point of diminishing return on the well's net present value.
Unconventional fracturing techniques, such as high-rate waterfracs, waterflooding, or steam stimulation, produced water and cuttings reinjection, CO2 sequestration, and coalbed methane stimulation, are difficult to model because of strong interactions among the fracturing process, geomechanical changes in the porous media, and reservoir fluid flow. The resulting strong poroelastic/thermoelastic effects, permeability and porosity changes, and possible rock failure make current conventional fracturing models inadequate in such circumstances. Therefore, it is necessary to develop new models that include all of these mechanisms and that are capable of conducting integrated data analysis.
This paper presents a new fracturing model with all of these mechanisms included. The model fully couples fracture mechanics with reservoir and geomechanics simulation. This methodology allows us to model fracture initiation and propagation, post-frac multiphase cleanup in the reservoir and fracture and pre-frac and post-frac well performance in a changing stress and pressure environment, all within the same system.
The model couples a 3D finite element geomechanics model with a conventional 3D finite difference reservoir flow simulator. The geomechanics module implicitly models fracture propagation via displacements on the fracture face. The flow and geomechanics/fracturing are coupled in an iterative manner that is equivalent to full coupling of geomechanical and reservoir flow modeling. The 3D (planar) fracture geometry and the pressures in it are the common dynamic boundary conditions for the flow and stress modules. The new iterative process yields smooth fracture propagation, and the model has been tested on classical fracturing problems.
A field example demonstrates the validity and advantages of the approach. To illustrate the model capabilities, we model a waterfrac stimulation performed in Bossier tight-gas sands. The model results show that the model is capable of matching complex injection history and calibrating the stress-dependency of formation permeability.
Nonparametric regression is a set of techniques for estimating a regression curve without making strong assumptions about the shape of the true regression function. These techniques are therefore useful for building and checking parametric models, as well as for data description. Kernel and nearest-neighbor regression estimators are local versions of univariate location estimators, and so they can readily be introduced to beginning students and consulting clients who are familiar with such summaries as the sample mean and median.
This paper discusses how to analyze past performance and predict futureperformance of tight gas wells stimulated by massive hydraulic fracturing (MHF)using finite fracture flow-capacity type curves. The limitations ofconventional pressure transient analysis and other methods of evaluating MHFtreatment are discussed. A set of constant well-rate and wellbore-pressure typecurves is presented.
Introduction
Because of the deteriorating gas supply situation in the U.S. and theincreasing demand for energy, the current trend is to consider seriously theexploitation and development of low-permeability gas reservoirs. This has beenpossible because of changes in the economic climate and advances in wellstimulation techniques, such as massive hydraulic fracturing (MHF). It nowappears that MHF is a proven technique for developing commercial wells inlow-permeability or "tight" gas formations. As the name implies, MHF isa hydraulic fracturing treatment applied on a massive scale, which may involvethe use of at least 50,000 to 500,000 gal treating fluid and 100,000 to 1million lb proppant. The purpose of MHF is to expose a large surface area ofthe low-permeability formation to flow into the wellbore. A low-permeabilityformation is defined here as one having an in-situ permeability of 0.1 md orless.
Methods for evaluating a conventional (small-volume) fracturing treatmentare available, but the evaluation of an MHF treatment has been a challenge forengineers. To evaluate the success of any type of fracture stimulation, prefracturing rates commonly are compared with postfracturing production rates.These comparisons are valid qualitatively if both pre- and postfracturing ratesare measured under similar conditions (that is, equal production time, samechoke sizes, minimal wellbore effects, etc.). Unfortunately, to evaluate thesuccess of different kinds of fracturing treatments, pre- and postfracturingproduction rates often are measured pre- and postfracturing production ratesoften are measured and compared using not only the same well tested underdissimilar conditions, but also the same kind of comparisons between differentwells that may even have different formation permeabilities. Thus, resultsoften are invalid and may cause misleading conclusions. Moreover, suchcomparisons do not help predict long-term performance. To predict long-termperformance for MHF wells, reliable estimates of fracture length, fracture flowcapacity, and formation permeability are needed.
Pressure transient methods for analyzing wells with small-volume fracturingtreatments are based on the concept of infinite or high fracture flow capacityand are used to determine the effectiveness of a stimulation by estimating thefracture length. Our experience indicates that these methods are not adequatefor analyzing wells with finite flow-capacity fractures. Such methods provideunrealistically short fracture lengths for MHF wells provide unrealisticallyshort fracture lengths for MHF wells with finite flow-capacity fractures.Furthermore, fracture flow capacities cannot be determined.
Includes associated paper SPE 8145, "Type Curves for Evaluation andPerformance Prediction of Low-Permeability Gas Wells Stimulated by MassiveHydraulic Fracturing."
Thesupport-vector network is a new learning machine for two-group classification problems. The machine conceptually implements the following idea: input vectors are non-linearly mapped to a very high-dimension feature space. In this feature space a linear decision surface is constructed. Special properties of the decision surface ensures high generalization ability of the learning machine. The idea behind the support-vector network was previously implemented for the restricted case where the training data can be separated without errors. We here extend this result to non-separable training data.
High generalization ability of support-vector networks utilizing polynomial input transformations is demonstrated. We also compare the performance of the support-vector network to various classical learning algorithms that all took part in a benchmark study of Optical Character Recognition.
Deep Convolutions for In-Depth Automated Rock Typing
- E E Baraboshkin
- L S Ismailova
- D M Orlov
Baraboshkin, E. E., Ismailova, L. S., Orlov, D. M. et al. 2020. Deep Convolutions for In-Depth Automated Rock Typing. Comput Geosci 135: 104330.
https://doi.org/10.1016/j.cageo.2019.104330.
Application of Machine Learning to Accidents Detection at Directional Drilling
- E Gurina
- N Klyuchnikov
- A Zaytsev
Gurina, E., Klyuchnikov, N., Zaytsev, A. et al. 2020. Application of Machine Learning to Accidents Detection at Directional Drilling. J Pet Sci Eng 184:
106519. https://doi.org/10.1016/j.petrol.2019.106519.
Overview of the Russian Oilfield Services Market-2019
- G Kamyshnikov
- A Kolpakov
Kamyshnikov, G. and Kolpakov, A. 2019. Overview of the Russian Oilfield Services Market-2019, https://www2.deloitte.com/content/dam/Deloitte/
ru/Documents/energy-resources/oil-gas-survey-2019-en.pdf (accessed 9 April 2021).
Prospects of Russian Oil Development: Life Under Sanctions
- T Mitrova
- Grushevenko
- A Malov
Mitrova, T., Grushevenko., and Malov, A. 2018. Prospects of Russian Oil Development: Life Under Sanctions. Report, SKOLKOVO Energy Centre
(SEneC), Moscow, Russia.
Reduced Order Reservoir Simulation with Neural-Network Based Hybrid Model. Paper presented at the SPE Russian Petroleum Technology Conference
- P Temirchev
- A Gubanova
- R Kostoev
Temirchev, P., Gubanova, A., Kostoev, R. et al. 2020a. Reduced Order Reservoir Simulation with Neural-Network Based Hybrid Model. Paper presented
at the SPE Russian Petroleum Technology Conference, Moscow, Russia, 22-24 October. SPE-196864-MS. https://doi.org/10.2118/196864-MS.
Deep Neural Networks Predicting Oil Movement in a Development Unit
- P Temirchev
- M Simonov
- R Kostoev
Temirchev, P., Simonov, M., Kostoev, R. et al. 2020b. Deep Neural Networks Predicting Oil Movement in a Development Unit. J Pet Sci Eng 184:
106513. https://doi.org/10.1016/j.petrol.2019.106513.