A statistical screening methodology is presented to address uncertainty related to main geological assumptions in green field modeling. The goals are the identification of the entire range of uncertainty on production, learning which are the most impacting geological uncertain inputs and understanding the relationships between geological scenarios and classes of dynamic behavior.
The paper presents the methodology and an example application to a green field case study. The method is applied on an ensemble of reservoir models created by combining geological parameters across their range of uncertainty. The ensemble of models is then simulated with a selected development strategy and dynamic responses are grouped in classes of outcome through clustering algorithms. Ensemble responses are visualized on a multidimensional stacking plot, as a function of the geological input, and the most influential parameters are identified by axes sorting on the plot. Geological scenarios are then classified on dynamic responses through classification tree algorithms. Finally, a representative set of models is selected from the geological scenarios.
The example study application shows a final oil recovery uncertainty range of 4:1, which is reasonable for a green field in lack of data. Such high range of uncertainty could hardly be found by common risk assessment based on fixed geological assumptions, which often tend to underestimate uncertainty on forecasts. Ensemble outcomes are grouped in four classes by oil recovery, plateau strength, produced water, and breakthrough time. The adoption of such clustering features gives a broad understanding of the reservoir dynamic response. The most influential geological inputs among the examined structural and sedimentological parameters in the example application result to be the fault orientation and channel fraction. This screening result highlights the main drivers of geological uncertainty and is useful for the following scenario classification phase. Classification of the geological scenarios leads to five classes of geological parameter sets, each linked to a main class of dynamic behavior, and finally to five representative models. These five models constitute an effective sampling of the geological uncertainty space which also captures the different types of dynamic response.
This paper will contribute to widen the engineering experience on the use of machine learning for risk analysis by presenting an application on a real field case study to explore the relationship between geological uncertainty and reservoir dynamic behavior.
SUMMARY We propose an automatic fault interpretation method by using convolutional neural networks (CNN). In this method, we construct a 7-layer CNN to first estimate fault orientations (dips and strikes) within small image patches that are extracted from a full seismic image. With the estimated fault orientations, we then construct anisotropic Gaussian functions that mainly extend along the estimated fault dips and strikes. We finally stack all the locally fault-oriented Gaussian functions to generate a fault probability image. Although trained by using only synthetic seismic images, the CNN model can accurately estimate fault orientations within real seismic images.
Mighani, Saied (Department of Earth Atmospheric and Planetary Sciences, Massachusetts Institute of Technology) | Lockner, David A. (U.S. Geological Survey, Menlo Park) | Kilgore, Brian D. (U.S. Geological Survey, Menlo Park) | Sheibani, Farrokh (Department of Earth Atmospheric and Planetary Sciences, Massachusetts Institute of Technology) | Evans, Brian (Department of Earth Atmospheric and Planetary Sciences, Massachusetts Institute of Technology)
Enhanced reservoir connectivity generally requires maximizing the intersection between hydraulic fracture (HF) and preexisting underground natural fractures (NF), while having the hydraulic fracture cross the natural fractures (and not arrest). We have studied the interaction between a hydraulic fracture and a polished saw-cut fault. The experiments include a hydraulic fracture initiating from a pressurized axial borehole (using water) that approaches a dry fault that is inclined at an angle θ with respect to the borehole axis. The experiments are conducted on Poly(methyl) Meta Acrylate (PMMA) and Solnhofen limestone, a finely grained (<5 μm grain), low permeability (<10 nD) carbonate. The confining pressure in all experiments is 5 MPa, while the differential stress (1-80 MPa) and approach angle, θ (30, 45, 60, 90°) are experimental variables. During the hydraulic fracture, acoustic emissions (AE), slip velocity, slip magnitude, stress drop and pore pressure are recorded at a 5 MHz sampling rate. A Doppler laser vibrometer measures piston velocity outside the pressure vessel to infer fault slip duration and a strain gauge adjacent to the saw-cut provides a near-field measure of axial stress.
For PMMA, the coefficient of friction was 0.30 and sliding was unstable (stick-slip). The approaching HF in PMMA created a tensile fracture detected by AE transducers ~100 μs before the significant stick-slip event (45% stress drop and slip velocity of ~60 mm/s) and was arrested by the fault at all fault orientations and differential stresses, even at 90° fault orientation and 80 MPa differential stress. For Solnhofen limestone, we observed stable sliding at a coefficient of friction of 0.12. In contrast to PMMA, the HF in Solnhofen consistently crossed to the other side of the fault. When the HF crossed the fault, it produced a small stress drop (<10%) and slip velocity of only 0.5 mm/s. Theoretical models by
We use here a fully hydraulically-mechanical coupled, 3-D model (Damjanac and Cundall, 2014) to simulate fault reactivation during a hydraulic fracturing treatment. Synthetic seismicity from the model helps quantify seismic energy released by the slippage on the fault. The model is based on a case study in the Horn River Basin by Snelling et al., 2013a. The multi-stage hydraulic fracture model is able to reproduce seismic deformation characteristics observed in field data. Results show that even stages distant from the fault have an influence on the slippage on the fault with a delayed effect. If the first injection stage is the closest to the fault, a large area will be slipping. Successive stages will have a lesser impact due to stress shadowing. If the first stage is farthest from the fault, then slippage on the fault will be gradual, reducing the amount of seismic moment release in a short period of time. This model can be used as a framework to examine the impact of other geomechanical characteristics or other operational factors, which could help establish best practices to mitigate seismicity when faults begin to be active.
Induced seismicity has become a concern for hydraulic fracturing operations in British Columbia and Alberta, Canada. Seismic monitoring is now mandatory for stimulation of two shale formations in this region. The challenge of hydraulic stimulations in areas prone to induced seismicity remains because mitigation can only be achieved with a good understanding of the underlying mechanisms linking multi-stage hydraulic fracturing operations and induced seismicity.
Geomechanical modeling is the best way to understand this link because it allows investigation of the interactions between multiple hydraulic fractures by modeling different injection scenarios and assessment of the sensitivity to different parameters. Many authors have proposed models to investigate induced seismicity (for instance, Goertz-Allmann and Wiemer, 2013; Rutqvist et al., 2013). Most find a strong correlation between pore pressure increase and areas where large magnitude events occur. The models indicate that the increase in pore pressure is caused by the hydraulic fracture following fluid injection.
None of these models can produce synthetic seismicity for quantitative comparison with recorded seismicity. The multi-stage hydraulic fracture model presented here is based on a fully hydraulically-mechanical coupled, 3-D model (Damjanac and Cundall, 2014) which produces synthetic seismicity, which can help quantify the seismic energy released by slippage on faults (Zhang et al., 2015).
A second process automatically extracts from those images fault surfaces represented by meshes of quadrilaterals. A third process uses differences between seismic image sample values alongside those fault surfaces to automatically estimate fault throw vectors. While some of the faults found in one 3D seismic image have an unusual conical shape, displays of unfaulted images illustrate the fidelity of the estimated fault surfaces and fault throw vectors. INTRODUCTION Fault surfaces like those shown in the closeup views of Figure 1 are an important aspect of subsurface geology that can be derived from seismic images. Therefore, various fault tracking methods, including those proposed by Pedersen et al. (2002, 2003), Admasu et al. (2006), Kadlec et al. (2008) and Kadlec (2011), have been developed to extract such surfaces. The fault throws shown in Figure 1 are important as well, as they enable correlation of subsurface properties across faults. Among methods developed to estimate fault throws are those described by Borgos et al. (2003), Aurnhammer and Tönnies (2005) and Admasu (2008).