Search

Coupled flow-geomechanics model is needed for investigating the stress change, rock-compaction behavior, and stress-dependent properties in many practical reservoir scenarios. However, the coupled model of large-scale simulation problems usually encounters large matrix system and high computational expenses, where the time stepping is a crucial factor for numerical stability and computational efficiency. Here, we report an adaptive time stepping with the modified local error method to reduce iteration time and improve the computational efficiency for the coupled flow-geomechanics modeling. Firstly, the iterative coupling approach with the fixed-stress method is introduced, where the flow and geomechanics equations are sequentially solved at each time step. Secondly, since updating geomechanics module consumes most of the computing time of the coupled system, the modified local error method is mainly used for geomechanics module to adapt the time step size based on the change of displacement. Finally, a synthetic two-dimensional coupled problem is established to apply the proposed adaptive time stepping approach, where the numerical results including the computational efficiency are compared with the results from regular iteratively coupled method and the fully coupled model. The sensitivity about the local error tolerance on the numerical results is also investigated. The geomechanical responses regarding mechanical loading is validated by comparing with the analytical solution of Terzaghi's consolidation problem. The numerical results of the two-dimensional coupled model show the modified local error method not only yields a higher-order solution for better accuracy, but also significantly reduces the iteration number and computational time. [ABSTRACT FROM AUTHOR]

Compositional simulation is often required for modeling complex oil recovery processes, such as CO 2 enhanced oil recovery ( Chen and Reynolds, 2016; Dai et al., 2014 ). Flash calculation using Equation of State (EOS) is very CPU expensive in compositional simulation. Conventional approaches for initial guess of K-values usually have several levels of limitations. This paper reports our research motivated to provide trustworthy initial guess for flash calculation and further improve flash performance in isothermal processes. Reservoir fluid composition changes are categorized into two different scenarios: (a) slightly varying or constant composition changes in depletion process; and (b) sharp varying composition in gas injection process. Scenario (a) is reasonably parameterized as a compositional space of K-value and vapor mole fraction varied with pressure. Considering the actual simulation of gas injection process, a compositional space is defined via introducing a key pseudo composition as a controlling variable in addition to pressure. Flash calculation is formulated through direct solution of those non-linear equations in phase-split calculation, and solved by combining successive substitution iteration (SSI) and Newton-Raphson (NR) method. Using interpolation from the constructed compositional space, it is found that the initial guess of K-values for flash calculation are always obtained with safeguard. A variety of fluid samples has been applied for testing this new flash technique. The results show that initial guesses of K-values based on Wilson’s correlation results cause much more flash iterations and even failures in some gas injection examples. On the other hand, the proposed method shows significant advantages in reducing the number of flash iterations in all examples. In addition, the SSI procedure in flash can even be bypassed and only NR procedure and a simple Rachford-Rice Equation Preconditioning is sufficient. [ABSTRACT FROM AUTHOR]

Conventional compositional simulators are usually difficult to interpret the different gas oil ratio (GOR) from tight oil reservoirs, and this also indicates an unreliable prediction of ultimate hydrocarbon recovery. We realize that there are two issues related to the compositional simulation of production in tight oil reservoirs. Firstly, tight oil reservoirs typically exhibit extremely small matrix pore size in the order of nanometers, so the capillary pressure between vapor and liquid phases is considerable such that the PVT of the confined fluid deviates from that of the bulk fluid with capillary pressure ignored. Secondly, during depletion process, rock compaction causes pore space reduction and brings remarkable changes in rock properties. In this work we implement rigorous confined fluid phase behavior calculation depending on capillary pressure and rock compaction in a fully compositional simulator. Capillary pressure in matrix nanopores is calculated by Leverett J-function. Further, the impact of capillarity on phase equilibrium is taken into account through modifying the stability test and two-phase flash calculation. Dynamic rock compaction is considered in the simulator via rock compaction tables, such that fluid mobility decreases with permeability reduction and capillary effect is simultaneously coupled. The unique implementation in the simulator captures the dynamic behavior of rock and fluid properties in tight oil reservoirs. Typical suppression of bubble point pressure and reduction of oil viscosity and density is observed from our simulation results. Reservoir-scale simulation results show that this model resolves the problem of the inconsistent GOR in tight oil production and greatly facilitates the history matching process. The enhanced compositional simulation will ultimately improve our understanding of tight oil reservoirs and provide better guidance for recovery prediction. [ABSTRACT FROM AUTHOR]

Shale gas reservoir is comprised of highly heterogeneous porosity systems including hydraulic/secondary fractures, inorganic and organic matrix. Multiple non-Darcy flow mechanisms in the shale matrix further bring challenges for modeling. In this paper, we developed a framework combining a multi-physics compositional simulator with Multi-Porosity Modeling preprocessor for gas storage and transport in shale. A Triple-Porosity Model is used to characterize the three porosity systems in shale gas reservoirs. In the fracture porosity the heterogeneous impact of secondary fractures distribution on matrix-to-fracture fluid transfer is revealed by shape factor distribution. They are upscaled with superior accuracy from a detailed Discrete Fracture Network Model (DFN) sector model, where orthogonal hydraulic fractures are explicitly discretized. With the occurrence of nano-pores in shale matrix, the interaction between pore-wall and gas molecules is considered via Knudsen diffusion and gas slippage. Gas adsorption on the pore-wall of organic matrix is modeled by extended Langmuir isotherm. The inter-porosity and intra-porosity connectivities in the Triple-Porosity Model are flexibly controlled by arbitrary connections. Our results show that gas production in the Triple-Porosity Model with shape factor upscaled from DFN exhibits different production performance from models with uniform shape factor distribution. The deviations are caused by the dominance of different regions at different production periods. Connection topology in the shale gas reservoir is also comprehensively assessed. We demonstrate that the intra-porosity connections in the inorganic and organic matrix have negligible impact on the global gas flux, while the inter-porosity connections have different levels of importance for the gas production. Moreover, different combinations of flow and storage mechanisms are investigated. We show that Langmuir desorption maintains reservoir pressure, but gas slippage and Knudsen diffusion accelerate the pressure drop. Both mechanisms contribute to improve the gas production and the consideration of them simultaneously improve gas production most. [ABSTRACT FROM AUTHOR]

Optimally drilling new wells in a developed reservoir is an essential strategy to potentially tap remaining oil for a complete life circle of oilfield development. Further, the determination of optimal infill drilling targets is a challenging issue which involves the integration of data, experts' knowledge and human decisions. The decision process can be essentially regarded as a systematic evaluation of drilling attractiveness. To automate drilling attractiveness evaluation, we develop a decision support engine using rule-based cognitive computing to rank and recommend drilling candidates. Such drilling candidates are chosen by the quantification of regional drilling attractiveness. Then we use two cases with different settings to show its general applicability and human-like reasoning abilities. The reasoning process considers expertise and human-involved uncertainties. The expertise is characterized by certain representation of fuzzy rules sets. Our results highlight the potential of our recommendation engine in pinpointing the most productive drilling location. And our method avoids the expensive reservoir simulation runs. Moreover, fuzzy drilling attractiveness evaluation can serve as an alternative initialization method of model-based infill well optimization, which avoids local optimum problem and greatly saves iteration time. Our approach extends human's reasoning capability and accelerates human's decision-making process with very low computational cost. • A framework for ranking and recommending infill drilling candidates is developed. • Expertise are quantified in rule-based infill drilling attractiveness evaluation. • The approach accelerates decision-making process with very low computational cost. • The method can serve as a proper initialization of model-based well optimizations. [ABSTRACT FROM AUTHOR]

Naturally fractured vuggy carbonate reservoirs play an important role in the world's oil and gas production. Pore-scale modeling is treated as an important approach for unveiling the flow mechanism in carbonate reservoirs. The realistic regeneration of fractured vuggy reservoirs needs to preserve the geological structures and complex pores morphologies. Nevertheless, regeneration of such reservoirs is a great challenge because of the co-existence of fractures and vugs over a wide range of scales. In this paper, the reference fracture-vug network will be divided into two systems: matrix-fracture and matrix-vug. The two systems are reconstructed in two ways: (I), the matrix-vug and matrix-fracture systems are generated by improved simulated annealing method (ISAM) and multiple point statistics algorithm (MPS), respectively; (II), both of the matrix-vug and matrix-fracture systems are generated by MPS. When applying for ISAM, the multipoint histogram mainly depends on the N-point template. The template is a unit configuration, which is considered as the characteristic of the geometric features. Based on this, we introduce the correlation coefficient matrix to determine an optimal template, and the accuracy of template selection algorithm is also discussed. The optimal template represents essential geological features of the training image, and its size is much smaller than the template size selected by entropy. Therefore, it can improve the computing efficiency while preserving the key properties of the network. Then, a superimposing method is proposed to create the fracture-vug network by combining the matrix-vug with matrix-fracture models. The superimposing method using ISAM (for matrix-vug model generation) and MPS (for matrix-fracture model generation) is called SMIM, while superimposing method using MPS (for both of matrix-vug and matrix-fracture models generation) is called SMM. Finally, the spatial connectivity, local percolation probability and patterns reproduction of the generated models are used to compare with those of the reference model. The results show that the 7-point template selected by the correlation coefficient matrix is enough to get the spatial features of the matrix-vug system. The generated fracture-vug network using the selected template is able to retrieve the patterns of training image (reference model) more accurately than other used templates. The fracture-vug network generated by superimposing method produces better results than the other methods (all named as “direct method”). The network simulated with SMIM algorithm matches the reference model better than SMM algorithm. The probable reason maybe exist in the different generation methods used for matrix-vug systems. The local percolation probability (LPP), treated as a general geometric characterization method, is used to characterize the connectivity of measured units at a given local porosity. The results of the probability analysis show that SMIM is better in depicting the expected patterns of geological heterogeneities and in producing higher degree of connectivity of pore space than the SMM. The pattern reproduction of SMIM model is assessed by scanning the generated model, and the ratio of pattern reproduction number to that of reference model (i.e. the similarity) is around 97.28%. These results may guide the further application of the proposed method to field-scale modeling of naturally fractured vuggy reservoirs. [ABSTRACT FROM AUTHOR]

We present an E nhanced D iscrete F racture N etwork ( EDFN ) model for multiphase flow simulation in fractured reservoirs. EDFN can efficiently represent complex fractured porous media and accurately simulate fluid exchange between matrix and fracture. By using EDFN the fracture network is discretized with minimum number of grids based on the fracture intersecting points and fracture extremities. The matrix is also optimally decomposed into coarse blocks with different geometries using a rapid image processing algorithm. Each coarse matrix block is designed to be associated with one fracture grid and then discretized logarithmically to handle 1D flow between matrix and fracture. EDFN can greatly optimize the discretization for fractured porous media to conform interconnected fractures of arbitrary orientations. Via unstructured format, EDFN is successfully linked with an in-house unstructured compositional simulator. We validate the accuracy of EDFN through a number of multiphase flow simulations in fractured porous media with/without considering capillary pressure. Its efficiency is demonstrated to be superior by using a much less grid blocks comparing with other approaches. Ultimately EDFN is applied to predict hydrocarbon production in a shale oil reservoir and it enables to capture multi-scale fluid transfer in such complex system. [ABSTRACT FROM AUTHOR]

This paper presents an approach called, E nhanced D iscrete F racture N etwork ( EDFN ) for fractured reservoirs modeling. The purpose of this approach is to efficiently represent complex fracture network and accurately simulate the fluid exchange between matrix and fracture. In this approach, fracture network is discretized with a minimum number of grids depending on the fracture intersecting points and fracture extremities. Besides, fracture network naturally decomposes matrix into coarse-scale block with variable geometries, and the partition is optimized by a rapid image processing algorithm. Each coarse matrix block is locally associated with a fracture grid, and it is equivalently discretized to rectangular fine-scale grid blocks. Fine-scale grid blocks are communicated with local fracture grids through one-dimensional flow transport. Therefore, the EDFN model maximally optimizes the discretization process for fractured reservoirs, and it is very appropriate for the simulation of reservoirs with arbitrarily oriented interconnected fractures. Through benchmark with different grid discretization approaches, the EDFN model provides very consistent results and its accuracy is validated. Besides, we also demonstrate that the EDFN model is able to simulate fractured reservoirs with a much smaller number of grid blocks than other approaches. Finally different non-Darcy flow mechanisms in shale gas reservoirs are incorporated into the EDFN model and the impact of those flow mechanisms are investigated. [ABSTRACT FROM AUTHOR]

We present an adaptive Multiscale Mixed Finite Element Method (MsMFEM) for modeling multiphase flow in highly heterogeneous reservoir. In this framework, fractional flow model is used to approximate pressure and velocity solution on coarse scale, while resolution in fine scale is honored by the basis functions, which are calculated by local problems on fine scale. The adaptive computation in the flow problem is permitted by two different basis functions. In the numerical examples, our adaptive MsMFEM is applied to solve two dimensional and three dimensional reservoir simulation problems with highly heterogeneous porosity and permeability fields. The preliminary numerical simulation results presented show a significant speedup in comparison with the reference numerical method, which encourage and is beneficial for further investigation of the proposed method for reservoir numerical simulation. [ABSTRACT FROM AUTHOR]