|Theme||Visible||Selectable||Appearance||Zoom Range (now: 0)|
Explaining Traditional Engineering Models It is a well-known fact that models of physical phenomena that are generated through mathematical equations can be explained. This is one of the main reasons behind the expectation of engineers and scientists that any potential model of the physical phenomena should be explainable. Explainability of the traditional models of physical phenomena is achieved through the solutions of the mathematical equations that are used to build the models. Explanations of such models are achieved through analytical solutions (for reasonably simple mathematical equations) or numerical solutions (for complex mathematical equations) of the mathematical equations. Solutions of the mathematical equations provide the opportunities to get answers to almost any question that might be asked from the model of the physical phenomena. Solutions of the mathematical equations are used to explain why and how certain results are generated by the model. It allows examination and explanation of the influence and effect of all the involved parameters (variables) on one another and on the model's results (output parameters).
Introduction Petroleum data analytics is a solid engineering application of data science in petroleum-engineering-related problems. The engineering application of data science is defined as the use of artificial intelligence and machine learning to model physical phenomena purely based on facts (e.g., field measurements and data). The main objective of this technology is the complete avoidance of assumptions, simplifications, preconceived notions, and biases. One of the major characteristics of petroleum data analytics is its incorporation of explainable artificial intelligence (XAI). While using actual field measurements as the main building blocks of modeling physical phenomena, petroleum data analytics incorporates several types of machine-learning algorithms, including artificial neural networks, fuzzy set theory, and evolutionary computing.
Horizontal wells in liquids-rich shale plays are now being drilled such that lateral and vertical distances between adjacent wells are significantly reduced. In multistacked reservoirs, fracture height and orientation from geomechanical effects coupled with natural fractures create additional complications; therefore, predicting well performance using numerical simulation becomes challenging. This paper describes numerical-simulation results from a three-well pad in a stacked liquids-rich reservoir (containing gas condensates) to understand the interaction between wells and production behavior. The reservoir simulator used for this study was designed to handle unstructured-grid-based simulation cases. Most of the numerical reservoir simulators that are used for modeling horizontal wells with multiple hydraulic fractures are based on structured grid cells in which the hydraulic fractures are modeled as symmetric biwing fractures perpendicular to the wellbore.
Summary Complex flow mechanisms, such as Knudsen diffusion, are encountered in the shale matrix because of the presence of nanopores. Numerous apparent‐permeability models have been proposed to capture the ensuing non‐Darcy flow behavior. However, these models are not readily available in most commercial reservoir simulators, and ignoring these mechanisms can potentially underestimate the overall matrix conductivity. This work implements an explicit coupling strategy for integrating a pressure‐dependent apparent‐permeability model in reservoir simulation. The numerical models are subsequently used to study the effects of apparent‐permeability modeling and natural‐fracture distribution on gas production and water loss during flowback. The effects of multiphase‐flow functions on fluid retention are also assessed. A set of 3D reservoir models are constructed using field data obtained from the Horn River shale‐gas reservoir. First, stochastic 3D discrete‐fracture‐network (DFN) models are scaled up into equivalent continuum dual‐porosity/dual‐permeability models. An apparent‐permeability (Kapp) model accounting for contributions of slip flow, Knudsen diffusion, and surface pore roughness is applied at each gridblock. A novel coupling scheme is formulated to facilitate the updating of Kapp after a certain specified time interval, capturing the pressure dependency of the Kapp. The sensitivity of the updating frequency is analyzed. The results reveal that incorporating these additional flow mechanisms by means of the apparent‐permeability formulation could potentially increase the overall gas‐production prediction by up to 11%, depending on the average pore radius, reservoir pressure, and several other matrix or fluid properties. The implications of Kapp modeling in water‐loss mechanisms are further examined through a set of sensitivity analyses, where the effects of multiphase‐flow functions and DFN distributions are systematically investigated. The following interesting findings are observed: Ignoring Kapp modeling could overestimate water recovery. Fracturing‐fluid propagation and long‐term water recovery are strongly affected by the secondary‐fracture intensity; increase in secondary‐fracture intensity would enhance water loss during flowback. Gas production is highly affected by the amount of water in the near‐well region. In a gas/water system, compressibility of the in‐situ fluids renders the effects of countercurrent imbibition and water retention to be more complex from those observed in similar water/oil systems. This work offers a novel, yet practical, scheme for representing the pressure‐dependent matrix apparent permeability in the flow simulation of shale reservoirs. The proposed method captures the non‐Darcy flow behavior caused by the complex transport mechanisms occurring in nanosized pores. Most importantly, this coupling procedure can be implemented in existing commercial reservoir‐simulation packages. The results have revealed a few interesting insights regarding the potential implications in fracturing design and estimation of stimulated reservoir volume.
Conventional plunger lifting is a transient process that consists of cyclic openings and closings of a gas well. Because of this complex behavior, using traditional physics-based models to simulate the coupled behavior of reservoir and wellbore performance is computationally rigorous and challenging. Therefore, this study proposes a machine learning-based approach to simulate gas production from plunger-lifted wells and facilitate the optimization of this process. The model is developed and validated using field data.
Typically, high-frequency (1 minute) measurements of plunger arrival time as well as casing and tubing pressure, are available in a plunger-lifted well. In addition, some wells are equipped with individual high-frequency measurements of gas flow rate. However, in most cases, there is a single gas flow rate meter available for the entire well pad. Therefore, a machine learning methodology is formulated with input variables that include plunger arrival time, tubing and casing pressure, and instantaneous gas flow rate as an output variable. Due to practical considerations regarding plunger lift operation, the approach assumes that a training set (one week) is smaller than a testing set (one month). A feed-forward neural network model is trained and is found to provide results with acceptable accuracy. The architecture of the network is obtained by performing a grid search and by minimizing a mean squared error. In the next step, obtained gas production is treated as a function of "on" (opening) and "off" (closing) time periods. The objective of the second model is to reproduce the data and to construct a response surface by varying "on" and "off" time periods.
Based on the results from several plunger-lifted gas wells, both models have a unified architecture that requires tuning weight coefficients with a training/development dataset. The neural network model to simulate the gas flow rate performs well; it is evaluated with common statistical parameters. The model requires gas flow rate measurements from routine production tests to build the training set. Having a gas flow rate model provides the opportunity to train another machine learning model as a function of "on" and "off" time periods. The new model is validated using the data during the final week of production history. The relative error between the data and the model is approximately 10%, which ensures the reliability of the model. A surface response is constructed over a range of "on" and "off" time periods to find an optimum point maximizing total gas production during the validation period (final week).
Optimization results demonstrate that "off" time (fall + buildup) should be minimized, and "on" time (upstroke + after-flow) should be at a certain threshold.
Current industry practice to optimize plunger lift cycles is based on factors such as average plunger rise velocity, and load factor. However, these methods do not optimize the actual variable of interest that is gas production. The unique contribution of the proposed approach is that it provides a robust tool to monitor the gas flow rate from an individual plunger-lifted well (flow rate allocation) and to optimize plunger lift cycles based on cumulative gas production. The model runs fast and can complement existing alarm systems on SCADA to adjust controller set-points in real-time.
The proposed method is a model using data-driven modeling. As many databases are available, we propose new alternatives to production analysis using exploratory data analysis (EDA) and deep neural network (DNN) techniques. It is an economical and time-saving model and workflow that can replace and alternate the traditional physics-based modeling. In this study, field data used about 1239 wells from Montney shale formation in Alberta, Canada. Through EDA, we verified the correlation between each variable of datasets and data distribution, and 1143 wells were used as training data for DNN model through data preprocessing such as outlier analysis, scatter plot, etc. The database used for the study was collected through database of GeoLOGIC systems ltd. The data here is largely divided into three features. In the case of well information, completion and fracturing data, production data and well information, there are true vertical depth, latitude, longitude, and well direction. total proppant placed volume. The production data used as the dependent variable in the DNN model is cumulative gas production for 12-month. Comprehensive machine learning techniques were applied to further improve predictive performance and prevent overfitting problem. First, we analyzed EDA and statistical analysis with various variables related to productivity. Second, developed model were designed and data preprocessing was performed to select input variables that are highly correlated with the output variables. Third, through variable importance analysis based on random forest (RF), gradient boosting tree (GBM), extreme gradient booting (XGBoost). We found a parameter that is highly correlated with cumulative gas production. Finally, we proposed the applicability of categorical variables. Then we performed hyperparameter optimization, a difficult problem for the DNN model. This paper is Not only is this paper intended to propose a robust model, but it also provides the insight that petroleum engineers can use as proxy models to replace complexity reservoir modeling or estimate approximate productivity of shale reservoirs before drilling.
Alqahtani, Fahd Mohamad (Norwegian University of Science and Technology (NTNU)) | Dahouk, Mohamad Majzoub (Whitson AS) | Whitson, Curtis H. (Norwegian University of Science and Technology (NTNU) / Whitson AS) | Chuparova, Ellie (Norwegian University of Science and Technology (NTNU))
This paper presents a comprehensive study of the influence of layer-wise fluid heterogeneity in the production performance of tight unconventional wells. Specifically, a reservoir simulation model is history-matched to actual production data from the field, using a black-oil reservoir simulator based on a planar-fracture symmetry element model. The history matched model is then used to illustrate the impact of multiple history-match "equivalents" that forecast GOR differently with uniform in-situ fluid assumption versus GOR variation in each parasequence (
This is the last paper of a series of studies and previous presentations (by the same authors) that have studied the impact of petrophysical and fluid heterogeneities on well GOR performance. The first paper uses a broad range of geoscientific studies to assess the possibility, and support the most probable fluid heterogeneity scenario in a tight unconventional – i.e. each parasequence in a well may undergo different in-situ hydrocarbon generation that leads to different in-situ compositions and layer-wise solution GORs. This paper is a field application and extension of that earlier work, focused on the impact on GOR performance over time and uncertainty in ultimate recovery (EUR).
We use field daily-metered production data and a black-oil PVT formulation (based on a basin-wide EOS model) to history match a 3D finite difference horizontal well model with a planar fracture symmetry element model. The history-matched model is used to study the impact of fluid heterogeneity on history match quality and forecasted GOR performance. The fluid heterogeneity is treated as a two-layered system with very limited communication between the layers except through the wellbore/hydraulic-fracture. Each layer represents a parasequence containing a uniform composition (i.e. in-situ solution GOR is different for each layer). The cases studied cover both (a) equal petrophysical properties in each layer with different initial solution GOR, and (b) different petrophysical properties k and φ for each layer with different solution GOR.
The study shows that several non-unique fluid initializations (rock and fluid heterogeneities) can lead to a comparable history match of producing GORs, while the forecasted GOR performance may vary according to the assumptions made for initial fluid distribution. It is also shown that even with known rock heterogeneity (i.e. known petrophysical properties for each layer), there may still be multiple fluid initializations that yield similar early-time producing OGR behavior, and sometimes similar OGR behavior throughout the historical data available.
The fluid sample data that – if available – could be used to narrow the degree of fluid heterogeneity in a well is seldom (if ever) available. The main consequence is that a history match of well GOR performance should be made using a range of plausible GOR fluid initializations – from uniform to layer-wise solution GOR variation (constrained by early GOR performance). The different history-matched models can then be used to forecast oil and gas rates that quantitatively capture the uncertainty of in-situ fluid distribution.
Rate forecasting of both oil/condensate and gas has been a long-standing challenge in tight unconventional resource plays. This paper gives a pragmatic approach to capture the uncertainty in initial fluid heterogeneity, and how this uncertainty would yield a range of forecasted GORs and EURs.
Xiaobing, Bian (Sinopec Research Institute of Petroleum Engineering) | Shidong, Ding (Sinopec Research Institute of Petroleum Engineering) | Tingxue, Jiang (Sinopec Research Institute of Petroleum Engineering) | Shuangming, Li (Sinopec Research Institute of Petroleum Engineering) | Haitao, Wang (Sinopec Research Institute of Petroleum Engineering) | Ran, Wei (Sinopec Research Institute of Petroleum Engineering) | Bo, Xiao (Sinopec Research Institute of Petroleum Engineering) | Yuan, Su (Sinopec Research Institute of Petroleum Engineering) | Guanyu, Zhong (Sinopec Research Institute of Petroleum Engineering) | Luo, Zuo (Sinopec Research Institute of Petroleum Engineering) | Jun, Zhou (Sinopec Research Institute of Petroleum Engineering)
ABSTRACT With streamlined and standardized treatment progress, the technology of well factory has been widely used in horizontal shale gas wells in the US, substantially improving multi-well fracturing efficiency. However, the technology is still at an early stage in China. It is of vital importance to set a pad as a separate unit to establish comprehensive optimization model of multi fracturing parameters, in order to get the optimal economic net present value (ENPV) of a pad. Then, multi parameters could be optimized simultaneously combining BP neural network model and genetic variation algorithm. Simulation results demonstrate that based on present geological and production condition in F shale gas play located in Sichuan Basin, smaller well spacing, more clusters, less fracture length and accordingly less fluid volume, will bring maximal effective stimulated reservoir volume (ESRV) and lower cost. Seventeen pad applications show that the hydraulic fracturing treatment period decreased by 30∼42%, with average well production increased by 51.2% in A block and 15.8% in B block respectively. 1. BACKGROUND Nowadays, the well factory hydraulic fracturing mode is prone to be adopted in the development of shale gas play (Li et al., 2011; Yan et al., 2015; Pope et al., 2009; Clarkson et al., 2012; William et al., 2010). In the US, it is streamlined and standardized for multiple wells fracturing simultaneously on a single well platform, greatly improved the hydraulic fracturing efficiency, production performance and fracturing fluids recycling (Cheng et al., 2012; Robert et al., 2012; Fonseca et al., 2013; Passman, 2018; Chiarandini, 2018). The well factory mode has a trend of decreasing well spacing (57 m), decreasing cluster spacing (5 m), and increasing cluster numbers (8∼10 clusters), as well as injecting less fracturing fluid per stage. Due to large number of wells on a platform, unit cost of a single well decreases with higher efficiency.
Wang, Qian (Henan Polytechnic University and The University of Queensland) | Donovan, Diane (The University of Queensland) | Reay, Thomas (The University of Queensland) | Thompson, Bevan (The University of Queensland) | Rodger, Iain (The University of Queensland Centre for Natural Gas) | Zhou, Fengde (Arrow Energy Ltd.) | Su, Xianbo (Henan Polytechnic University, and Collaborative Innovation Center of Coalbed Methane and Shale Gas for Central Plains Economic Region) | Yazici, Emine Sule (Koç University)
This paper investigated the impact of geological and engineering factors on coal seam gas production in horizontal wells. The results were then used to compare the performance of proxy models based on linear and quadratic response surfaces and Universal Kriging (UK) to models based on Polynomial Chaos Expansion (PCE). A simple reservoir model was created using a commercial reservoir modelling software package which includes the capability to construct proxy models. The simple model was used for the dynamic modelling of cumulative gas production and peak gas rate under uncertainty in the input variables, e.g. the reservoir principal permeability and permeability in orthogonal directions varied as ratios of the principal permeability, porosity, gas content, coal saturation and the angle between the horizontal well and the principal permeability direction. The simulation results for cumulative gas production and peak gas rate were used to generate training data for proxy models, which were then used to predict the simulated output for other combinations of input parameters. Error analysis was conducted for each proxy model and used to compare and contrast the different modelling techniques. In addition, we investigated the sensitivity of the model to changes in the input variables. The results indicate that, for the given study, proxy models based on linear regression were not good estimators for cumulative gas production and peak gas rate. However, when enough training points were utilized, all other techniques provided good estimates. The best performing proxy model was a cubic PCE. In addition, the cubic PCE proxy model provided direct access to the sensitivity of the gas production to changes in the values of the input variables. Considering the main effects, the changes in the principle permeability and the porosity were predominant factors for the cumulative gas production and peak gas rate, respectively. As for the pairwise interactions, the combined effect of the principle permeability and the coal saturation had the most impact on the cumulative gas production, followed by the combined effect of the drilling angle and the ratio of the permeabilities in the y-direction and the x-direction, indicating that gas production can be improved by optimizing the orientation of the horizontal well. In addition, the combined effect of the coal saturation and the porosity had the most impact on the peak gas rate, an interesting result that warrants further investigation in future studies.
Summary Production from liquids–rich shale reservoirs in the US and Canada has increased significantly during the past few years. However, a rigorous understanding of shale rocks and fluid flow through them is still limited and remains a challenge. Thus, the objective of our research is developing a 3D physics–based model for simulating fluid flow through these types of multiporosity rocks. This is important given the recent spread of these types of reservoirs throughout the world. Simulation of liquids–rich shale reservoirs is performed with the construction of an original fully implicit 3D multiphase modified black–oil finite–difference numerical formulation, which uses a multiporosity approach as well as diffusion from solid kerogen. The multiporosity system includes adsorbed porosity, organic porosity, inorganic porosity, natural–fracture porosity, and hydraulic–fracture porosity. A numerical model is developed with capabilities to handle dissolved gas in the solid part of the organic matter, adsorption/desorption from the organic pore walls, viscous– and non–Darcy–flow mechanisms (slip flow and Knudsen diffusion), and stress–dependent properties of natural and hydraulic fractures. Examples of simulated results are presented as crossplots of pressure, production rates, and cumulative production vs. time. These plots are used to show the contributions of free gas, adsorbed gas, and dissolved gas to fluid production from liquids–rich shale reservoirs. Results indicate that both desorption and gas diffusion positively affect shale performance. Simulation results demonstrate that not taking into account desorption and diffusion from solid kerogen leads to underestimating production from liquids–rich shale reservoirs. Furthermore, the simulation study shows that long periods of time are required for the effects of these two mechanisms to be manifested. This helps to explain why shales have been produced over long periods of time (several decades), such as in the case of Devonian wells in the Appalachian Basin. The type of 3D simulation model for multiporosity liquids–rich shale reservoirs developed in this paper is not currently available in the literature. The approach implemented in this paper provides a novel and important foundation for simulating complex shale reservoirs.