Gong, Yiwen (The Ohio State University) | Mehana, Mohamed (Earth and Environmental Sciences Division, Los Alamos National Laboratory, Los Alamos, USA) | Xiong, Fengyang (The Ohio State University) | Xu, Feng (Research Institute of Petroleum Exploration and Development CO., LTD, CNPC) | El-Monier, Ilham (China National Oil and Gas Exploration and Development Corporation)
Rock elastic moduli are one of the major perspectives for the hydraulic fracturing design. Among all of them, Young's modulus and Poisson's ratio essentially control fracture aperture for the proppant placement. The objective of this work is to predict the elastic moduli by applying data mining techniques as a comparison to the experimental measurements. We have collected attributes representing the pore structure, mineralogy and geomechanical characteristics. We implemented classification techniques such as k-means, hierarchical and PAM (partition around medoids). PAM results in more evenly-distributed clusters compared to the rest. Artificial Neural Network (ANN) is used for regression. We formulated two scenarios; firstly, all the data is grouped into one group and the other involves performing the regression on the clustered data. Interestingly, both scenarios yield acceptable results. The classification results could guide the fracturing operations where clusters with high brittleness, low anisotropy and high microfracture intensity could be identified as fracture candidates. Still the main limitation to unleash the machine learning capabilities in this domain is the data scarcity
Objectives/Scope: In order to maximize the recovery of hydrocarbons from liquids rich shale reservoir systems, the cause and effect relationships between production and the stimulation methods need to be clearly understood. In this study, we utilize multivariate regression models to narrow down the variables in flow simulation models and their range. We then use the flow simulation model to understand the fractured well production behavior and field wide well performance in a liquids rich petroleum system in the Duvernay Basin.
Methods, Procedures, Process: Statistical models assume no physical relationship between the model parameters and the response variable, which in this case is produced volumes over a period of time. On the other hand, simulation studies incorporate physical mechanisms of flow to model and predict the production behavior. The simulation models, however, fall short of incorporating all the mechanisms contributing to the production behavior in the complex shale gas reservoir. Thus there is a need for integration of statistical approaches of understanding production behavior along with physics based model and simulation approach. We use the statistical methods to identify the important physical mechanisms that control the production.
Results, Observations, Conclusions: Multivariate linear regression analysis of the 6 month produced volume and its relationship with parameters such as fracture fluid volumes used, proppant weight placed, number of stages fractured provides a model with reasonably good correlation. The 6 month produced volumes correlate with large proppant weights, lower fluid placements and greater density of fracture stages. Use of Random Forests machine learning algorithm on the dataset confirms that the total proppant placed, well length completed with fractures have high importance coefficients. In order to examine the well performance using full physical models, fractured well simulations are performed on particular wells using the trilinear model. The trilinear model predictions are then compared against other production analyses and the regression model results for consistency. The models showed that in the absence of stress dependent permeability, the production forecast was much higher. Thus, stress dependent permeability appears to be an important factor in the modeling and prediction of production from liquids rich shale reservoirs.
Novel/Additive Information: In this study we describe a method to understand the production data from a liquids rich shale reservoir, by integrating multivariate linear regression analysis, machine learning algorithms along with physical model simulations. The results are novel and offer a method to validate either approach to understand cause and effect relationships. This approach may be classified as a new hybrid modeling workflow that may potentially be used to optimize stimulation techniques in liquids rich shale reservoirs.
Booking reserves for unconventional, multi-frac wells is a critical business process, but to be done effectively, often requires significant time investment and multiple interpretation techniques. Although reserves can be estimated quickly with decline curve analysis (DCA) alone, the subjectivity in DCA makes it challenging for evaluators to estimate reserves with appropriate levels of uncertainty and maintain consistency between evaluators. The objective of this paper is to present a fast, systematic, yet rigorous methodology for estimating 1P, 2P and 3P estimated ultimate recoveries (EURs) for new wells. This methodology utilizes regression to correlate easy to obtain, early life indicators of well performance to 2P EURs, which have been estimated from more detailed interpretations. Multiple methodologies are presented for estimating 1P and 3P EURs.
Guo, Qingbin (PetroChina Tarim Oilfield Company) | Qiu, Bin (PetroChina Tarim Oilfield Company) | Zhao, Yuanliang (PetroChina Tarim Oilfield Company) | Fan, Zhaoya (Schlumberger) | Chen, Jichao (Schlumberger) | Han, Yifu (Schlumberger) | Zhang, Tao (Schlumberger) | Li, Kaixuan (Schlumberger) | Yu, Hua (Schlumberger) | Jiang, Lei (Schlumberger) | Wei, Guo (Schlumberger) | Yu, Daiguo (Schlumberger)
The Kuqa foreland thrust belt, as a secondary tectonic unit of the Tarim basin at the front of the Tianshan Mountains, is a foreland basin that formed in the Late Tertiary. The lower Cretaceous Bashijiqike tight sandstone in the basin is an ultralow-permeability and low-porosity reservoir. The Kuqa foreland thrust belt includes Kela, Keshen, Bozi, Zhongqiu, and Alvart blocks. Although these blocks developed under the same sedimentary conditions, the permeability-porosity relationship and wireline log response can be very different among the blocks. Whereas the shallow zone has been had E&P activities for decades, fully understanding the fluid properties, the porosity-permeability relationship, and distribution pattern of gas in the deep to ultradeep zone is of strategic significance and can provide the experience for the exploration of similar gas reservoirs in China and worldwide. The main target zone depth varies from 6000 m to 8000 m, and the formation pressure is near or exceeds 20,000 psi. Compared to a time-consuming and costly drillstem test (DST) operation, the wireline formation test (WFT) is the most efficient and cost-saving method to confirm hydrocarbon presence. However, the success rate of WFT sampling operations in the deep Kuqa formation is less than 50% overall, mostly due to the formation tightness exceeding the capability of the tools. Therefore, development of an optimized WFT suitable to the formation was critical.
More than 30 WFT wells in Kuqa foreland thrust belt were studied to understand the well and formation conditions causing the success or failure of these WFT operations. By doing a statistical analysis of more than 1000 pressure test points, we researched the relationship between mobility and petrophysical logs such as neutron, density, gamma ray, resistivity, P-sonic, etc. Several statistical mathematic methods were applied during this study, including univariate linear regression (ULR), multiple linear regression (MLR), neural network regression analysis (NNA), and decision tree analysis (DTA) methods. A systematic workflow was formed to mine data information, and we delivered a standard chart of the relationship between mobility and the petrophysical logs, an integrated equation based on MLR, and an NNA model that can be applied to WFT feasibility analysis.
These methods can be considered the foundation of artificial intelligence (AI), which can be used in future mobility automatic prediction. This provides a rough estimation of the mobility and sampling success rate and enables WFT optimization to be conducted in advance.
Zhang, Yingchun (CNOOC Research Institute Co., Ltd.) | Xu, Wei (CNOOC Research Institute Co., Ltd.) | Zou, Jingyun (CNOOC Research Institute Co., Ltd.) | Jing, Zhiyi (CNOOC Research Institute Co., Ltd.) | Fang, Lei (CNOOC Research Institute Co., Ltd.) | Liu, Jun (CNOOC International Limited)
In complex clastic reservoirs, deviation often exists in oil saturation derived from logging interpretation due to the borehole conditions and log quality. Especially in thin-sand reservoirs, oil saturation is generally lower than actual results because of boundary effect. An innovative approach of saturation height function coupled with rocktype is provided to improve the accuracy of saturation prediction in well logs and spatial distribution. The model results are compared with log derived results.
The new approach is based on the routine and special core analysis of over 100 core samples from the complex clastic reservoir in the north of Albert Basin in Uganda. Discrete rocktypes (DRT) are determined by flow zone index and pore throat radius which indicate the fluid flows. After converting the capillary pressure (Pc) data to reservoir conditions, Lambda curve fitting (Sw = A * PcB + C) is used to fit each capillary pressure curve. Then, a robust relationship between the fitting coefficients (A, B, C) and rock properties (i.e. porosity and permeability) is expressed as a nonlinear function for each DRT. Combined with the height above free water level, a water saturation (Sw) model is constructed by SHF within DRT model.
Using the porosity and permeability obtainedfrom routine core analysis, FZI and pore throat radius are calculated (e.g., by Winland function). Five different rocktypes (DRT1-5) are defined in the delta sand reservoir in the north of Albert Basin with distinct pore textures. The distinguishment is in accordance with the shape of capillary pressure curve, that is, the flow capability increases from DRT1 to DRT5. A strong correlation between Pc and Sw processed by Lambda curve is acquired for each core sample. Meanwhile, 3 coefficients A, B and C can be obtained in Lambda formula. By nonlinear regression, coherent relation between each factor and reservoir properties (porosity and permeability) for each DRT are obtained. Height above the free water level is estimated by geometrical modeling on the oil water contact. The Sw model is constructed by the new SHF function coupled with DRT model. It showed that the water saturation derived from SHF is highly consistent with log derived results and NMR results. Moreover, it provides more precise results in thinner sands and in spatial distribution.
Based on the identified different rocktype, a new SHF derived from capillary pressure data is utilized to establish the relationship between saturation, the height above the free water level and rock properties. The approach can significantly improve the accuracy of saturation prediction of thin reservoir and reasonably depict the spatial distribution characteristics of saturation. Furthermore, the approach will provide a more precise result in hydrocarbon volume calculation and numerical simulation.
Annular Pressure Build-up (APB) is caused by heating of the trapped drilling fluids (during production) which may lead to burst/collapse of the casing or axial ballooning, especially in subsea HP/HT wells. The objective of present paper is to apply machine-learning tools to increase precision of the APB estimation, and thereby improve the fluid and casing design for APB mitigation in a given well.
The APB estimation methods in literature involve theoretical and computational tools that accommodate two separate effects: volumetric expansion (PVT response) of the annulus drilling fluids and circumferential expansion (and corresponding mechanical equilibrium) of the well casings. In the present work, machine-learning algorithms were used to accurately model ‘fluid density=
The study demonstrated that, in several subsea scenarios, a relatively small error in the experimental fluid PVT data itself can lead to significant variation of in APB estimation. The machine-learning based models for ‘density =
Reservoir fluid characterization is critical to understanding the nature and phase behavior of reservoir fluids. This process has typically been undertaken using laboratory analyses, a time-intensive and costly process which also provides compositional data. Over time, correlations have been developed to predict the PVT properties of crude oil based on parameters such as solution gas-oil ratio, saturation pressure, viscosity, and density. These correlations have had shortcomings such as utilizing a leave-one-out approach, or recently, focused on non-inferable methods such as Neural Networks. This work utilizes compositional data, hitherto neglected in PVT correlations, as input into an inferable machine learning algorithm which can be used to predict PVT properties of crude oil from the Niger Delta basin.
Data containing bubble point pressure, solution gas-oil ratio, and oil formation volume factor alongside composition were obtained and used to develop models. Machine learning model training techniques such as data preprocessing, transformation and hyper-parameter tuning were undertaken. The elastic net regression algorithm utilizing a cross-validation approach was used to develop the models. This ensured an adequate bias-variance tradeoff.
The resulting models were compared with established correlations such as Standing & Katz. Upon statistical analyses performed comparably. The bubble point pressure model, solution gas-oil ratio, oil formation volume factor achieved R-squared value of 0.87, 0.95 and 0.84 respectively on the validation dataset. The models are expressed in the form of equations which can be used in petroleum engineering calculations or implemented in reservoir simulation software. By implementing this approach, a framework for utilizing machine learning for Petroleum Engineering problems which produces inferable results is established. Given potential discoveries in the Niger Delta, upon obtaining compositional data, these set of equations can be used to predict the reservoir crude oil PVT properties, leading to savings in time, cost, and effort, while obtaining actionable and accurate results.
Hydrocarbon production from shale formation has become an essential part of the global energy supply in the past decade. The life of a project in an unconventional play significantly depends on the prediction of Estimated Ultimate Recovery (EUR). However, the conventional methodology to predict EUR becomes less accurate for shale formations, which significantly affects the economics returns of projects in unconventional plays. The objective of this article is to investigate the most important independent variables, including petrophysics and completion parameters, to estimate EUR by the machine learning algorithm. A novel machine learning model based on Random Forest Regression is introduced to predict EUR and to rank the importance of the independent variables.
In this article, production/petrophysics/engineering/ data with more than 25 variables from 4000 wells in Eagle Ford is summarized for analysis. The data is collected from production monitoring, well logging, well testing, seismic interpretation and lab experiments. This paper has three major components. Firstly, a multivariate linear regression model is created to predict the overall EUR. Secondly, the spatial autocorrelation analysis is carried out to identify whether spatial variables could affect the accuracy of the multivariate regression model. Thirdly, the Random Forest Regression models are trained to examine their reliability in predicting EUR with spatially autocorrelated data. The importance of key predictors is also identified. The final models are tuned with optimized hyperparameters. Through the article, the predictive capabilities of each Random Forest Regression model are discussed in detail to understand the physics behind unconventional hydrocarbon production mechanisms.
The results and workflow presented in this paper are insightful and novel. Firstly, we test the multivariate regression analysis with all the petrophysics and completion variables using the backward elimination method. This widely used model has a limitation of excluding the spatial information. In order to identify the impact of spatial variable, we calculate the Moran's Index and find out that the data in this study is clustered or spatially autocorrelated. The p-value for EUR, Oil EUR and Gas EUR are 0.000002, 0.000000 and 0.12, which all reject the null hypothesis that the data is randomly distributed. To include the spatial information in the prediction, we use advanced machine learning technology, Random Forest, to predict the EUR with a combination of petrophysics, completion variables and spatial information. The key variables to predict EUR, Oil EUR and Gas EUR by the Random Forest Regression are identified. However, the importance of the key variables to predict Oil EUR and Gas EUR are different. Therefore, we split the overall EUR Random Forest Regression model (57% explained) into two prediction models, one for Oil EUR prediction and one for Gas EUR prediction. The Gas EUR Random Forest Regression model has better performance (76% explained) compared to the Oil EUR Random Forest Regression model (60% explained).
This study provides a deeper understanding of unconventional hydrocarbon production prediction from a big data perspective, and proposes a novel and reliable machine-learning model to predict EUR to evaluate economic returns in Eagle Ford. Compared to the traditional multivariate regression model, our Random Forest Regression models are more reliable. In addition, the Random Forest technique is able to rank the importance of the relevant independent variables, and the rank of importance can be applied to guide and to improve data collection and model training for further study on this topic. The workflow presented in this article can be also used to train data for other unconventional resource plays.
Data from seismic to production is integrated to build models to provide estimations of parameters such as petroleum volumetrics, pressure behavior, and production performance (
Reservoir dynamic simulation is the most applied process that integrates all reservoir data, where an Equation of State (EOS) is coupled with the objective to estimate the fluid thermodynamic state at each computational step. The simulation consists of iterative mathematical computations in which the reservoir-defined conditions at the previous time step is an input to determine the properties at the next and subsequent time steps. The calculated pressure is a fundamental variable in each time step, which means that a representative and high level of confidence Pressure Volume Temperature (PVT) model is required to avoid scale-up of errors resulting from fluid pressure estimation.
A PVT modeling includes three main stages: Fluid sample and data acquisition Laboratory analysis and fluid characterization The EOS model.
Fluid sample and data acquisition
Laboratory analysis and fluid characterization
The EOS model.
The emphasis in this work is on the EOS model, which is the fluid model used for the simulation process. The objective of this work is to analyze the main uncertainties associated with typical EOS modeling and defining the level of confidence of these EOS approaches. In this work, some of the most-used approaches for EOS modeling are reviewed. An assessment of these methods is also provided based on their application to actual petroleum fluids with the objective of defining their statistical level of confidence.
First, the study analyzes the sources of critical uncertainties in a PVT EOS model. Second, a statistical number of PVT laboratory studies of petroleum fluids is used to determine the level of confidence of four approaches that are based on the two well-known Peng-Robinson and Soave-Redlich-Kwong EOS. Third, statistical analysis is performed to determine the level of confidence of the different methods. Fourth, a correlation to determine the optimal number of pseudo-components is defined. These steps include: Characterization of fluid and heavy components Tuning Lumping.
Characterization of fluid and heavy components
As a result of this study, one can conclude: The level of confidence of the four analyzed approaches The significance of the difference between the analyzed methods A correlation to determine the optimal number of pseudo-components.
The level of confidence of the four analyzed approaches
The significance of the difference between the analyzed methods
A correlation to determine the optimal number of pseudo-components.
In this work, a statistical analysis over some of the most-used EOS modeling approaches and on a set of petroleum fluid PVTs was performed to determine the level of confidence of four EOS modeling methods. In addition, a correlation was introduced for
Monte Carlo simulation is a process of running a model numerous times with a random selection from the input distributions for each variable. The results of these numerous scenarios can give you a "most likely" case, along with a statistical distribution to understand the risk or uncertainty involved. Computer programs make it easy to run thousands of random samplings quickly. Monte Carlo simulation begins with a model, often built in a spreadsheet, having input distributions and output functions of the inputs. The following description is drawn largely from Murtha.