|Theme||Visible||Selectable||Appearance||Zoom Range (now: 0)|
Tran, Ngoc Lam (University of Oklahoma) | Gupta, Ishank (University of Oklahoma) | Devegowda, Deepak (University of Oklahoma) | Jayaram, Vikram (Pioneer Natural Resources) | Karami, Hamidreza (University of Oklahoma) | Rai, Chandra (University of Oklahoma) | Sondergeld, Carl H. (University of Oklahoma)
Summary In this study, we demonstrate the application of an interpretable (or explainable) machine‐learning workflow using surface drilling data to identify fracturable, brittle, and productive rock intervals along horizontal laterals in the Marcellus Shale. The results are supported by a thorough model‐agnostic interpretation of the input/output relationships to make the model explainable to users. The methodology described here can easily be generalized to real‐time processing of surface drilling data for optimal landing of laterals, placing of fracture stages, optimizing production, and minimizing fracture hits. In practice, this information is rarely available in real time and requires tedious and time‐consuming processing of logs (including image logs), core, microseismic data, and fiber‐optic‐sensor data to provide post‐job validation of fracture and well placement. Post‐completion analyses are generally too late for corrective action, leading to wells with a low probability of success and increasing risk of fracture hits. Our workflow involves identifying geomechanical facies from core and well‐log data. We verify that the geomechanical facies derived using core and well‐log data have characteristically different brittleness, fracturability, and production characteristics. We test and investigate several different supervised classifiers to relate surface drilling data to the geomechanical facies. The data were divided into training and test data sets, with supervised classification techniques being able to accurately predict the geomechanical facies with 75% accuracy on the test data set. The clusters predicted on test well (unseen data) were qualitatively verified using the microseismic interpretation. The use of Shapley additive explanations (SHAP) helps explain the predictive models, rank the importance of various inputs in the prediction of the facies, and provides both local and global sensitivities. Our study demonstrates that pre‐existing natural‐fracture networks control both the hydraulic‐fracture geometry as well as the production. Natural fractures promote the formation of complex fracture networks with shorter half‐lengths, which increase well productivity while minimizing fracture hits and neighboring‐well interactions. The natural‐fracture network is itself controlled by the geomechanical properties of the rock. The ability of the surface drilling data to reliably predict the geomechanical rock facies provides a powerful tool for real‐time optimization of wellbore trajectory and completions.
Unconventional reservoirs, such as Permian Basin, have fundamentally different production behaviors than that of the conventional reservoirs because of the low permeability of formations away from the stimulated volume. Thus, it is difficult to run full-field reservoir simulation to generate a full-field development plan. Even though satisfactory history matching for completion and production data can be achieved for wells at one location, it is difficult to directly apply the results to other areas in the same region. This is especially true in complex thick pay zone reservoirs, such as Permian Basin, where the complex geology, geomechanical properties, and resource properties all make the solution of frac hit (optimal well spacing) very challenging. To our best knowledge, this study is the first to integrate the detailed physics-based simulation, including fracturing simulation, coupled reservoir and geomechanics simulation, with machine learning (ML) to generate a sound workflow for well spacing optimization. In this workflow, a large field is first divided into several representative regions according to geology, geomechanical properties, and reservoir properties; and a typical well is selected for each region. High-quality physics simulation (including fracturing simulation, coupled reservoir simulation and geomechanics simulation) and history matching are then performed for a pair of parent and child wells. Additional well completion scenarios are built upon the base case, which serve as input to the following ML study. Various ensemble regression methods are applied to generate production predictions for unexplored reservoir locations in this field.
In recent years, many fracture simulation models have been developed to represent the complex geomechanical processes involved in hydraulic fracturing (F. Ajisafe, Shan, Alimahomed, Lati, & Ejofodomi, 2017; Morales et al., 2016; Pankaj, 2018b). Among them, well interference or well spacing optimization is a critical issue to solve in energy sector, especially during the current industrial downturn (F. O. Ajisafe, Solovyeva, Morales, Ejofodomi, & Porcu, 2017; Pankaj, 2018a; L. Wang & Yu, 2019; M. Wang, Wang, Zhou, & Yu, 2019). Detailed mechanistic study of integrated fracturing simulation and reservoir simulation have made significant progress during the recent years, which greatly helps to unlock unconventional resources and assist the industry to achieve economic goals (Alimahomed et al., 2017; Lashgari, Sun, Zhang, Pope, & Lake, 2019; Min, Sen, Ji, & Sullivan, 2018; Rodriguez, 2019). Nevertheless, when applying these technologies in unconventional field development and production, major uncertainties remain, including geology aspects such as stress orientation, stress anisotropy and natural fracture distribution, and completion aspects such as discrepancy between different completion strategies (Pankaj, Shukla, Kavousi, & Carr, 2018; Xiong, Liu, Feng, Liu, & Yue, 2019). A common concern is that even though current physics-based modeling and simulation could match the completion and historical production of a single well or multiple wells, it is still difficult to successfully transfer or scale up one small-area’s knowledge and experience to another area, because of the complexity and uncertainty of unconventional reservoirs (Xiong, Ramanathan, & Nguyen, 2019; Yeh et al., 2018).
Rezaei, Ali (University of Houston) | Siddiqui, Fahd (University of Houston) | Dindoruk, Birol (University of Houston / Shell International Exploration and Production Inc.) | Soliman, M. Y. (University of Houston)
In order to maximize the profitability of a well and minimize the cost, three key questions must be answered before drilling a well: Where to drill the well? What completion design is to be used? Which fluid type will be produced from the reservoir? These questions must be answered under the premise of maximizing profitability. In this study, we combine the recently developed artificial neural network (ANN) model with a global sensitivity analysis method to present a reduced-order model for addressing these questions.
We developed ANN models to predict the oil and gas production of the first year. The input of the model are parameters such as longitude, latitude, true vertical depth, lateral length, fracturing fluid volume, proppant volume, and fracture stages. Next, we use the Sobol global sensitivity analysis to identify the dominant input variables and their interactions on the variation of the oil and gas production. Finally, we develop reduced-order models that can be represented as a simple algebraic expression consisting of simple mathematical functions. These equations can then be used to predict the production in the Eagle Ford shale rapidly by engineers on the field.
The ANN model used in this study predicted the oil and gas production of the first year with reasonable accuracy. Our model suggests increasing the number of fracture stages and proppant volume in the oil-bearing region. The suggestions for the gas bearing cases were opposite to the oil case. The Sobol global sensitivity approach used in this study captures the variation of the output parameters of the ANN model with respect to the changes in the input parameters. Also, it identifies the combined output variation due to the changes of multiple input parameters. After ranking the dominant contributing input parameters, the model was used to present a simple function to predict the oil and gas production of the first year (combined oil and gas). The function has the advantage to be used in a simple excel sheet and can rapidly predict the results. We compared the accuracy of the proposed reduced order model against the developed ANN model, and results showed less than 5% error in predictions.
For the first time, we have combined the data science methods with analysis of variance (ANOVA) based methods. This has resulted in a simple mathematical function to rapidly and directly predict the oil and gas from Eagle Ford shale, based on the input parameters that can be selected before drilling the well. Using the presented methodology, other such functions can be created for other shale plays and will aid engineers and decision-makers for field development to make reliable and quick decisions.
This study uses a machine learning framework to systematically analyze field production and completion data to understand the impact of frac-hits on parent and child wells and predict well spacing and completions design. Frac hits are one of the most pressing reservoir management issue that can enhance or compromise production over either the short-term or have sustained impacts over longer times. The extent of the impact is dictated by a complex interplay of petrophysical properties (high-perm streaks, mineralogy, etc.), geomechanical properties (near-field and far-field stresses, brittleness, etc.), completion parameters (stage length, cluster spacing, pumping rate, fluid and proppant amount, etc.) and development decisions (well spacing, well scheduling, etc.). As a result, the impact of frac-hits is not straightforward and difficult to predict.
The study uses data from the Meramec, Woodford and Wolfcamp formations. We develop an automated machine-learning based frac-hit detection algorithm that also quantifies the impact on the parent and child wells using matched decline curve models. We analyze about 500 parent and over 1100 child wells in the three formations. Our results show that the key factors governing the extent of the impact are the extent of depletion and producing oil rate of the parent well before frac hit, completion design parameters (fluid and proppant amount) and well spacing. Our machine learning analysis generates regression models to predict the impact of frac hits. These regression models are coupled with economic analysis to determine optimum spacing for any given completion design or optimum completion design for any given spacing.
The parent wells in all three formations had both positive and negative impact of the frac hits. Around 60–67% parent wells were negatively impacted while 33–40% wells were positively impacted. For the child wells, 71–85% wells were negatively impacted and 15–29% of the wells were positively impacted. Combining the impact on parent and child wells, the impact is dominated by the child wells as 69 to 82% of the parent-child pairs were negatively impacted and only 18–31% of the pairs were positively impacted. Considering percent loss in cumulative oil volumes in the next 5-years, in the Meramec, parent wells on average show a 16% reduction while child wells show a 39% reduction due to frac hits. The corresponding numbers for the Woodford formation are 19% and 37% and Wolfcamp formation are 20% and 22%, respectively. This translates to a parent well losing on average 40–50 thousand bbls in next five years and a child well losing on average 130–150 thousand bbls in the same period.
This study systematically analyzes available data to understand the impact of frac hits and formulates a machine learning-based well spacing-well completions matrix workflow that can easily be extended to other formations by integrating commonly available production and completions data.
Bian, Changrong (Sinopec Exploration & Production Research Institute) | Zhang, Dianwei (Sinopec Exploration & Production Research Institute) | Shen, Feng (GeoReservoir Research) | Wo, Yujin (Sinopec Exploration & Production Research Institute) | Sun, Wei (Sinopec Exploration & Production Research Institute) | Li, Jingliang (GeoReservoir Research) | Han, Juan (GeoReservoir Research) | Li, Shuiquan (GeoReservoir Research) | Ma, Qiang (Sinopec Exploration & Production Research Institute)
Delineating geometry of natural fractures realistically and understanding fracture stress sensitivity help to optimize well placement and well spacing design in shale gas reservoirs. This paper presents a methodology for building 3D hybrid discrete natural fracture network (DFN) models and using an analytical model to assess reactivation potential of natural fracture in the Longmaxi shale, Sichuan Basin. Small-throw faults and natural fractures ranging from seismic scale to well scale in shale reservoirs have important effects on the success of horizontal drilling and hydraulic fracturing. Seismic geometric multi-attributes at different resolution scales are used to classify seismic facies according to the degree of fracturing. Small-throw faults are delineated using seismic facies and validated against drilling data. We develop a discrete natural fracture network (DFN) model at the seismic scale by meshing fracture lineaments tracked from an enhanced curvature attribute. Fracture topologies are used for fracture connectivity analysis to build local fracture networks along and around the horizontal wellbores. Diffuse fractures at the small scale are modeled with curvature attributes and well data analysis under the constraint of the seismic facies. The analytical model incorporates fracture properties and geomechanical model to describe the deformation of natural fractures due to hydraulic fracturing. Fracture stress-sensitivity are assessed based on changes of fracture volumes under different stress conditions. Characterized reactivated local fracture networks at different scales along the horizontal wells are used to map out volumetric extent of zones with potential to develop tensile and shear deformation during hydraulic fracturing. Available microseismic data from the hydraulic fracture stimulation of the reservoir is used to validate the fracture models. Our stress sensitivity analysis indicates that reactivation potential of natural fractures varies considerably, mainly depending on natural fracture size and orientation, rock mechanical properties and anisotropy of horizontal stresses. DFN models reveal that fracture concentrations are correlative with the footprint of observed microseismic events. Comparison of 3D natural fracture models with the microseismic event distribution shows that vertical variation of fracture properties in the laminated shale reservoir adds complexity for fracture propagation. A case study is used to illustrate the efficiency of the methodology. Fracture models at different scales and associated fracture stress-sensitivity can be used as a predictive tool for locating new wells and completion design in shale gas reservoirs.
Xue, Xu (Texas A&M University) | Yang, Changdong (Texas A&M University) | Park, Jaeyoung (Texas A&M University) | Sharma, Vishal Kumar (Texas A&M University) | Datta-Gupta, Akhil (Texas A&M University) | King, Michael J. (Texas A&M University)
Summary Multistage hydraulically fractured horizontal wells provide an effective means to exploit unconventional reservoirs. The current industry practice in the interpretation of field response often uses empirical decline‐curve analysis or pressure‐transient analysis/rate‐transient analysis (PTA/RTA) for characterization of these reservoirs and fractures. These analytical tools depend on simplifying assumptions and do not provide a detailed description of the evolving reservoir‐drainage volume accessed from a well. Understanding of the transient‐drainage volume is essential for unconventional‐reservoir and fracture assessment and optimization. In our previous study (Yang et al. 2015), we developed a “data‐driven” methodology for the production rate and pressure analysis of shale‐gas and shale‐oil reservoirs. There are no underlying assumptions of fracture geometry, reservoir homogeneity, and flow regimes in the method proposed in our previous study. This approach depends on the high‐frequency asymptotic solution of the diffusivity equation in heterogeneous reservoirs. It allows us to determine the well‐drainage volume and the instantaneous recovery ratio (IRR), which is the ratio of the produced volume to the drainage volume, directly from the production data. In addition, a new w(τ) plot has been proposed to provide better insight into the depletion mechanisms and the fracture geometry. w(τ) is the derivative of pore volume with respect to τ. In this paper, we build upon our previous approach to propose a novel diagnostic tool for the interpretation of the characteristics of (potentially) complex fracture systems and drainage volume. We have used the w(τ) and IRR plots for the identification of characteristic signatures that imply complex fracture geometry, formation linear flow, partial reservoir completions, and fracture‐interference/compaction effects during production. The w(τ) analysis gives us the fracture surface area and formation diffusivity, while the IRR analysis provides additional information on fracture conductivity. In addition, quantitative analysis is conducted using the novel w(τ) plot to interpret fracture‐interference time, formation permeability, total fracture surface area, and stimulated reservoir volume (SRV). The major advantages of this current approach are the model‐free analysis without assuming planar fractures, homogeneous formation properties, and specific flow regimes. In addition, the w(τ) plot captures high‐resolution flow patterns not observed in traditional PTA/RTA analysis. The analysis leads to a simple and intuitive understanding of the transient‐drainage volume and fracture conductivity. The results of the analysis are useful for hydraulic‐fracturing‐design optimization and matrix‐ and fracture‐parameter estimation.
Abstract Eagle Ford shale in South Texas is a major oil and gas production play of in the US Gulf Coast region. While some attribute the successful well performance of Eagle Ford to the technology advancement such as horizontal drilling and hydraulic fracturing, others credit the role of geological settings. However, it is still unclear what the individual or combined effects from these two sides are. Data-driven approaches, including Partial Least Square (PLS), Random Forest (RF), and Deep Neural Network (DNN), reveal relationships among the production, geological settings, and completion strategies. In this study, we considered six-month cumulative oil production as the well performance criterion for horizontal wells completed from 2015 to 2017. We selected completion parameters such as perforation length, proppant loading, and fluid volume. We selected structural depth, lower Eagle Ford Shale thickness, total organic carbon (TOC), number of limestone beds, and average bed thickness as the key geological controls on regional production. We calculated Spearman correlation coefficients to detect correlated input parameters and applied Singular Value Decomposition (SVD) to identify redundant input parameters. Then we performed partial linear square (PLS) regression to predict the six-month oil production from geological and completion parameters. We then used random forest (RF) and deep neural network (DNN) as non-linear machine learning techniques to predict six-month oil production and compared the prediction accuracies for these techniques against the recorded well performance using the coefficient of determination and mean squared error as criteria. Last, we ranked the relative importance of each input parameter using RF and Minimum Redundancy Maximum Relevance (MRMR). This paper first provides the rational of input variables selection. Then the construed model helps understand the effects of completion designs and geological variables on well productivity in the Eagle Ford. This might provide valuable information to help to make decisions for new well development. This concept can be generalized among other plays.
Abstract Researchers from both industry and academia have studied the tight oil resources intensively in the past decade since the successfully development of Bakken Shale and Eagle Ford Shale and made tremendous progress. It has been recognized that locating the sweet spots in the regionally pervasive plays is of utter significance. However, we are still struggling to determine whether the dominant control on shale well productivity is geologic or technical. Given certain geological properties, what is the best completion strategy? Most of the previous studies either analyze the completion data alone or divide the entire play into different data clusters by map coordinates and depth, which may neglect the heterogeneity in thickness and reservoir quality parameters. In our study, we first conducted stratigraphic and petrophysical analyses, using the regional variation in depth, thickness, porosity, and water saturation to capture the regional heterogeneity in the Bakken Shale Petroleum System. We selected approximately 2,000 horizontal wells targeting Mid Bakken Formation with detailed completion records and initial production dates during 2013 and 2014. Completion data inputs include normalized stage length, stage counts, normalized volume of fluid, and normalized volume of proppant. We investigated the relative importance of the geological and completion features exerts on the first-year production. Then we built the neural network model to identify the relationship between the first-year oil production with the important features. We separated the data into three sets for training, validation, and testing. After we trained the model using the training and validation set, we tested the model to estimate its robustness. Then, with given geological input, we provided the completion strategy that optimizes the first-year production. The developed technique provides a method to identify the best well location, understand the effectiveness of the completion strategy, and predict the well production. While the data used came from wells in the Bakken shale, the methodology applies in a similar way to other tight oil plays.
Abstract Interactions between reservoir rocks and engineering fluids during hydraulic fracturing have potential for modification of pores and flow paths by mineral transformation and precipitation and contamination of production fluids by metals and other salts. In order to infer how these interactions affect hydraulic fracturing performance, this study investigates interactions between engineering water and reservoir rocks from different chemofacies in Lower Eagle Ford Group under relevant temperature and pressure conditions. Rock samples were selected from five chemofacies in Lower Eagle Ford Group, which were divided by variations in different element concentrations and total organic carbon (TOC). Static and dynamic experiments were conducted at different condition. In static experiments, crushed reservoir rock samples were exposed to deionized water for three weeks at room condition. In the dynamic experiment, deionized water was continuously injected to the experimental system for three hours at reservoir condition. Rock samples were characterized by XRF before experiments to estimate major and trace elemental concentrations. Water samples after experiments were analyzed for ion contents, total dissolved solids (TDS), particle size and zeta potential. ANOVA single factor analysis using Tukey HSD and principal component analysis (PCA) were used to assess the similarity and difference in interactions between chemofacies. Water parameters were used to determine the tendency of suspensions to precipitate and potential to modify flow pathway during hydraulic fracturing. This study provides information on interactions likely forming between water and Eagle Ford reservoir rocks and key geochemical tracers indicative of the effective surface areas where interactions occur. The information enhances the understanding of water-rock interaction mechanisms and distribution of fracture networks.
Abstract Today there are still many challenges in the quantitative interpretation of downhole distributed temperature measurements to diagnose multistage fracture treatments in horizontal wells. These challenges include handling enormous amount of data measured by the sensors continuously in time and space domain, a readily-to-be-used fast but robust forward model to simulate temperature behavior, and an efficient algorithm to inverse the parameters that are of interest. Because multistage fracturing involves many uncertain parameters; ranging from reservoir properties to treatment design, to fracture geometries and conductivity; the problem is extremely complex when inverse the measured temperature to a downhole flow profile. This study presents an approach to combine forward and inversion models to interpret downhole temperature data. The goal is to improve computational efficiency. Examples use the field data from a gas well in Marcellus shale formation to illustrate the feasibility of quantitative interpretation of temperature measurement for fracture diagnosis. The forward model uses the fast marching method. The forward simulation is order-of-magnitude faster than even the semi-analytical model, which is the essential contribution to apply the method in the field case successfully. The inversion procedure starts with a sensitivity study to select the inversion parameters among various parameters such as fracture half-length, fracture conductivity, and determine the impact of their uncertainty on inversion. The inversion model uses the initial analysis on temperature gradient to identify the zones with significant temperature changes for interpretation and eliminates the rest of the data from interpretation. Thus, we obtain a prior estimation of the selected inversion parameters, which will be used as an initial guess of the inversion process. This prior estimation saves significant computation. The inversion is performed fracture by fracture either using parallel computing or sequential computing based on the sensor locations. We first show a synthetic example with multiple fractures to illustrate the approach, test the procedure accuracy and computation speed. The primary inversion parameter is flow rate, and either fracture length or fracture conductivity, with all other parameters as an additional constraint. With an adequate initial guess, the inverted parameters match the reference "true value" properly. The inversion process converges with limited iterations for each fracture. The operation time highlights the advantages of the inversion model. The guided initial guess ensures the gradient inversion method converge and avoid local minimization. Finally, a field application is performed using this inversion model and shows encouraging results. The results of the paper illustrate the feasibility and procedure of using temperature date to diagnose multistage fracture treatment. The proposed inversion model is fast and reliable which provides a promising tool that can be used to interpret downhole temperature data quantitatively.