|Theme||Visible||Selectable||Appearance||Zoom Range (now: 0)|
High-fidelity 3D engineering simulations are valuable in making decisions, but they can be cost-prohibitive and require significant amounts of time to execute. The integration of deep-learning neural networks with computational fluid dynamics may help accelerate the simulation process. Reducing a separation system’s footprint while increasing separation efficiency is demonstrated in an Oklahoma field trial. Reliable separation is becoming an enabling technology to help develop remote location resources and more difficult applications, such as heavy oil, produced water, sand disposal, and back-produced fluids in enhanced oil recovery. This paper provides details of comprehensive computational-fluid-dynamics (CFD) -based studies performed to overcome the separation inefficiencies experienced in a large-scale three-phase separator.
In this study, the authors investigated a fully data-driven approach using artificial neural networks (ANNs) for real-time virtual flowmetering and back-allocation in production wells. The authors present a new data-driven approach to estimate the injection rate in all noninstrumented wells in a large waterflooding operation accurately.
Measuring the flow of water, mud, and cuttings from a well is critical, and difficult. A new flowmeter design that promises to be both accurate and durable is one of three technologies featured in a JPT series on drilling measurement innovation. This paper describes a virtual metering tool that can monitor well performance and estimate production rates using real-time data and analytical models, integrating commercial software with an optimization algorithm that combines production and reservoir information. In this study, the authors investigated a fully data-driven approach using artificial neural networks (ANNs) for real-time virtual flowmetering and back-allocation in production wells. Australian technology developer MezurX is touting its newly introduced flow, density, and mud monitoring system as a significantly better alternative to the widely used Coriolis meter.
A real-time deep-learning model is proposed to classify the volume of cuttings from a shale shaker on an offshore drilling rig by analyzing the real-time monitoring video stream. The best way to know how drilling affects drill bits is to visualize the bits. A device that creates high-resolution images for precise measurements is one of three technologies being featured in a JPT series on drilling measurement innovations. For the upstream industry, where improvement in efficiency or production can drive significant financial results, there is no question that the size of the digital prize is huge. A group of people who really care how drillers code the memos added to the daily drilling report is the data scientists—who find that the coded tags do not match the activity.
The course introduction will attempt to answer the question: How will A.I. change the way we work in the Oil and Gas industry in the coming years? Looking at what is underway in other industries and guessing what type of projects are under development in R&D departments in our industry will help answer that question. Oil and Gas examples will be presented corresponding to each of the terms A.I., Machine Learning, and Deep Learning, allowing participants to reach a clear understanding on how they differ. The course will then focus on Deep Learning (DL) and address all key aspects of developing and applying the technology to Oil and Gas projects. What is DL and how different is it from traditional neural networks?
In oil and gas industry it is crucial to have reliable information on well, reservoir and boundary types and properties. Detailed information can be extracted from a proper interpretation of pressure and rate transients of well testing data. Though, there are times that even with an in-depth pressure transient analysis, a unique solution on well, boundary and especially the reservoir types cannot be obtained and makes it difficult or even impossible to extract correct information. In this study deep learning (DL) is used to tackle this problem by differentiate possible reservoir models and select the most appropriate model based on pressure derivative response. Accuracy of the classification model on real field data with known models is also explored.
Reservoir models can be identified by measuring the downhole pressure data and analyzing the changes in trends in pressure curves and especially pressure derivative curves. In this study, different DL algorithms are used to identify the basic characteristics of pressure derivative curves to determine reservoir model. Several possible well/reservoir/boundary types are considered to select the best model that can be used for well/reservoir/boundary property estimation. Before feeding the networks, training data curves would be shrunk in size using wavelet transform (WT) which is able to sustain the pressure derivative features in a much-compressed form to accelerate algorithm training and testing.
The technique used in this work is a time-efficient process that learns important signatures of pressure derivative curves to classify reservoir models. Unlike the conventional well testing methods in which models are determined from the visual inspection of the pressure and pressure derivative plots, the technique used in this study was trained with a dataset consists of hundreds of reservoir models generated by solving diffusivity equation under different well, reservoir, and boundary conditions. The procedure was applied to multiple field examples with known reservoir model and reservoir properties and proved the consistency and flexibility of the methodology for true reservoir model selection. DL-based models also shown to be very handy with excellent computational efficiency especially when dealing with the complex patterns on the pressure derivative curves. The study showed that the method has great capability to classify pressure derivative and can also tolerate noise when applied on real pressure data.
Large dataset used in this study can increase the comprehensiveness of the training and test data sets. The big advantage of the DL-based approach was the improvement in the pattern recognition of the pressure derivative curves without the need of any feature handcrafting or any prior knowledge of well, reservoir, and boundary types. ML proved to be a reliable, fast, and accurate technique that can significantly improve the process of well, reservoir, and boundary type detection based on pressure derivative curves.
Facies classification is significant for characterization and evaluation of a reservoir because the distribution of facies has an important impact on reservoir modelling which is important for decision making and maximizing return. Facies classification using data from sources such as wells and outcrop cannot capture all reservoir characterization in the inter-well region and therefore as an alternative approach, seismic facies classification schemes have to be applied to reduce the uncertainties in the reservoir model. In this research, a machine learning neural network was introduced to predict the lithology required for building a full field earth model for carbonate reservoirs in Sothern Iraq.
In the present research, multilayer feed forward network (MLFN) and probabilistic neural network (PNN) were undertaken to classify facies and its distribution. The well log that was used for litho-facies classification is based on a porosity log. The spatial distribution of litho-facies was validated carefully using core data. Once successfully trained, final results show that PNN technique classified the carbonate reservoir into four facies, while the MLFN presented two facies. The final results on a blind well, show that PNN technique has the best performance on facies classification. These observations implied this reservoir consists of a wide range of lithology and porotype fluctuations due to the impact of depositional environment.
The work and the methodology provide a significant improvement of the facies classification and revealed the capability of probabilistic neural network technique when tested against the neural network. Therefore, it proved to be very successful as developed for facies classification in carbonate rock types in the Middle East and similar heterogeneous carbonate reservoirs.
Ebadi, Mohammad (Skolkovo Institute of Science and Technology) | Bezyan, Yashar (Concordia University) | Zabihifar, Seyed Hassan (Bauman Moscow State Technical University) | Koroteev, Dmitry (Skolkovo Institute of Science and Technology)
The reservoir simulation is based on the solving of second-order nonlinear Partial Differential Equations (PDEs). Following the high-level of nonlinearity or irregular boundaries, analytical solutions are not applicable to solve the supposed PDEs. To numerically solve the PDEs, applying nonlinear solvers are recommended. Dependencies on derivatives and proper initial guesses are the main disadvantages of classic solvers. To overcome the mentioned obstacles, solving supposed equations based on Adaptive Neural Network (ANN) has been introduced.
The algorithm starts by introducing an initial set into the Nonlinear Simultaneous Algebraic Equations (NSAE). The outputs are compared with the desired matrix of zeros to generate the required error. The calculated vectors of errors and its derivation are firstly employed to update the ANN weights through applying the adaption laws, and secondly, create the input vector to run the ANN. The outputs of the ANN are considered as corrections to be made to the initial set. Then, the corrected initial set is reintroduced into equations. The procedure continues iteratively until the outputs of equations meet the required level of accuracy.
By taking advantages of the adaptive laws, the outputs of the presented algorithm have successfully been matched with answers of the classic solvers, but with less computational costs. The convergence of the shown algorithm has practically been examined by assuming various mathematical types of initial sets. The implemented algorithm has been robust enough to converge for different forms of the initial sets, even for invalid values like minus numbers. However, records indicate that the convergence rates are strongly dependent on the values of initial sets. Following the sensitivity analysis over the primary model of ANN lead to the optimized network, which could solve the supposed NSAE three times faster. It has been interpreted that the number of neurons (NN), the diagonal coefficient matrix of error (
In contrast to Newton's method as the most well-known nonlinear solver, the launched algorithm does not require any proper initial guesses. Moreover, the absolute independence of computing the partial derivatives of the Jacobian matrix and its inversion, which causes a notable reduction of computational costs, is the other remarkable advantage of the proposed approach. The represented algorithm can be taken as the platform to develop the next generation of simulators working based on machine learning.
Germik, a mature heavy oil field in Southeast Turkey, has been producing for more than 60 years with a significant decline in pressure and oil production. To predict future performance of this reservoir and explore possible enhanced oil recovery (EOR) scenarios for a better pressure maintenance and improved recovery, generation of a representative dynamic model is required. To address this need, an integrated approach is presented herein for characterization, modeling and history matching of the highly heterogeneous, naturally fractured carbonate reservoir spanning a long production history.
Hydraulic flow unit (HFU) determination is adopted instead of the lithofacies model, not only to introduce more complexity for representing the variances among flow units, but also to establish a higher correlation between porosity and permeability. By means of artificial intelligence (AI), existing wireline logs are used to delineate HFUs in uncored intervals and wells, which is then distributed to the model through stochastic geostatistical methods. A permeability model is subsequently built based on the spatial distribution of HFUs, and different sets of capillary pressures and relative permeability curves are incorporated for each rock type.
The dynamic model is calibrated against the historical production and pressure data through assisted history matching. Uncertain parameters that have the largest impact on the quality of the history match are oil-water contact, aquifer size and strength, horizontal permeability, ratio of vertical to horizontal permeability, capillary pressure and relative permeability curves, which are efficiently and systematically optimized through evolution strategy. Identification and distribution of the hydraulic units complemented with artificial neural networks (ANN) provide a better description of flow zones and a higher confidence permeability model. This reduces uncertainties associated with reservoir characterization and facilitates calibration of the dynamic model. Results obtained from the study show that the history matched simulation model may be used with confidence for testing and optimizing future EOR schemes.
This paper brings a novel approach to permeability and HFU determination based on artificial intelligence, which is especially helpful for addressing uncertainties inherent in highly complex, heterogeneous carbonate reservoirs with limited data. The adopted technique facilitates the calibration of the dynamic model and improves the quality of the history match by providing a better reservoir description through flow unit distinction.