Natural fractures have a dramatic impact on reservoirs in terms of oil recovery because they often control the hydraulic flow as conductors (open fractures) or barriers (sealed fractures). However, fracture parameters are poorly constrained by reservoir data, due to the low seismic resolution and to the clustering of 1D data along wells. To enhance flow prediction, we suggest improving the characterization of Naturally Fractured Reservoirs by integrating well data with the history of stresses obtained by three-dimensional structural restoration (3D balanced unfolding). When unfolding a layer, boundary conditions are applied to mesh displacements to unfold the upper horizon and remove the latest brittle and ductile deformation increment. A linear isotropic elastic model governs restoration behavior, accounting for mechanical contrasts in the reservoir. Three-dimensional strains and stresses are derived from these displacements. Orientation of theoretical tectonic fractures is then geomechanically deduced from the principal strain directions. Moreover, the orientations of fractures observed at wells (cores or image logs) provide implicit information on the principal directions of paleo-stress. Considering a given failure criterion, the observed fracture orientation is used to constrain the relative directions of paleo-stress components. Through this original approach, the 3D deformation is calibrated by observed geological data. Therefore, the hybrid geometrical and geomechanical restoration that has been developed accounts for reservoir heterogeneities and is globally constrained by paleo-stresses deduced from fractures observed along wells. This methodology is applied to a real reservoir located in North America and the strains predicted are used to generate 3D Discrete Fracture Networks. The key benefit of this approach, as compared to conventional methods, is to enhance fracture characterization by directly integrating observed fracture data into the geomechanical process. Introduction Modeling natural fractures has recently become a high priority issue for oil reservoirs because of their major impact on fluid flow. Because they constitute either barriers or preferential ways to hydraulic flow and thus greatly modify the permeability field, their characteristic parameters (location, orientation, size and aperture) must be evaluated over the whole studied domain. However, none of these attributes are well constrained by available subsurface data neither by well nor by seismic data. This article focuses on a specific parameter: fracture orientation, its integration and prediction into the modeling workflow. Joint orientations are usually described by stochastic and probability distribution laws estimated from well data. In the literature, the common approach adopted to predict orientation uses intensively geostatistical methods[1,2,3]. However, when the physical laws at the origin of a phenomenon are known, geostatistical approaches are not realistic: limited to near-well sampling, fracture statistics are biaised away from them. Therefore, predicting the orientation of joints based on their geomechanical origin enables significant reductions in uncertainty by using all the available static data to constrain the model at a field scale. The knowledge of the stress history over the whole studied domain is mandatory to predict realistic fracture orientations. A pioneering attempt of such prediction is described by Fischer and Wilkerson[4]. They forecast joint orientation from curvature analysis which can be related to fracturing in some cases. However, there is no mathematical evidence that fracture orientation should be correlated with curvature[4]. Other authors have already predicted fracture orientation based on the stress field evaluated with forward modeling method[5,6,7]. Based on a boundary element method (BEM), the stress field computed is directly linked to the geometry of faults but not to the geometry of horizons. However, horizons are indirectly accounted for the stress field computation by the choice of the mechanical model but away from faults, their geometry is disregarded in the modeling. The mechanical behavior of rocks is assumed to be constant over the whole subsurface domain which is not realistic. Moreover, in non-faulted reservoirs (such as the case study of this article), this methodology cannot be applied.

Adequate estimation of resources and reserves is critical for any prospect or field in the oil and gas industry. We propose improving the reliability of this estimation, first, by exploring the range of uncertainties through the use of scenario-based geological models; second, by ranking all generated models based on both static and dynamic responses; and, finally, by selecting few models representative of the uncertainty affecting the reservoir. In any field, the unknowns largely outnumber the known data, and, typically, some of these known data are biased due to preferential sampling. The classical approach controls the range of uncertainties based on multiple equiprobable realizations generated by using different random paths and by randomly sampling the data range. Our method improves that approach by considering additional realistic geological scenarios, which allows expanding the uncertainty space to sample from. A large number of models is generated as a result. Performing flow simulation on a large range of models would be computationally expensive. The common industry practice is to select models based on static volumetric measurements only. Our approach not only accounts for static volumes, but also adds computation of the flow-based connectivity using streamline simulation to account for the dynamic behavior of the models. A sensitivity analysis allows preserving only the most influential variables and discarding the negligible ones. Finally, static and dynamic responses from the remaining models are analyzed to identify scenarios with consistent low, mid, and high values for both responses. This results in the selection of few representative models that provide a greater assurance of recovering the probability distribution than would have been achieved by following common industry practices. Our methodology involves first enriching the uncertainty space by adding a dimension of geological scenarios. Then, it uses both static and dynamic responses to select a handful of representative models while still allowing adequate estimation of resources and reserves.

No abstract

Spatial distribution and proportions of facies in a depositional system are essential drivers of reservoir heterogeneity and connectivity and, thus, flow performance. Modeling facies relationship in complex depositional systems is one of the major challenges, particularly thin, elongated and sinuous bodies such as channels due to their high aspect ratio. Traditionally, such geological bodies have been generated using object models but with limitations in regards to the geometries of the associated facies that can be handled. In this paper, we propose a method to model a complex channelized depositional system using a multipoint statistics (MPS) method that efficiently integrates a non-stationary conceptual model with hard data and secondary information. The conceptual model is a clastic shallow-marine depositional system with fluvial and deltaic deposits. It comprises a transitional facies model with lateral variations and a transgressive-regressive sequence. The transgressive phase is a wave-dominated estuary environment, and the regressive phase comprises wave/fluvial-dominated delta deposits. The proposed methodology allows us to quickly create a 3D geocellular reservoir model of a complex depositional system that efficiently incorporates measured data and geological conceptual knowledge. The resulting MPS model satisfactorily reproduces facies patterns, proportions, and connectivity.

No abstract