GEOMECHANICALLY COUPLED SIMULATION OF FLOW IN FRACTURED RESERVOIRS

Transcription

1 PROCEEDINGS, Thirty-Eighth Workshop on Geothermal Reservoir Engineering Stanford University, Stanford, California, February 11-13, 2013 SGP-TR-198 GEOMECHANICALLY COUPLED SIMULATION OF FLOW IN FRACTURED RESERVOIRS C. A. Barton 1, D. Moos 1, L. Hartley 2, S. Baxter 2, L. Foulquier 1, H. Holl 3, R. Hogarth 3 1 Baker Hughes GMI Geomechanics Services Palo Alto, CA, 94306, USA colleen.barton@bakerhughes.com 2 AMEC Oxfordshire, OX11 0QB, UK 3 Geodynamics, Limited Brisbane, AU ABSTRACT The vast majority of geothermal reservoirs are naturally fractured at various intensities and scales. These fractures are usually the main pathways for fluid flow and thus their properties control to a large degree both the production profile from producing wells and the well injectivity. Because most fractures are stress-sensitive, their hydraulic conductivities will change with changes in bottom-hole-flowing and reservoir pressures causing variations in production profiles between wells. More specifically, fractures can hydraulically (partially) open or close due to a decrease (caused by injection) or increase (caused by production) in the effective stress. Because flow properties are a function of effective fracture aperture it is possible to predict reservoir behavior using the relationship between the mechanical behavior of natural fractures (in response to in situ stress and pore pressure changes) and their hydraulic properties. We discuss here a case study of the integration of geological, geophysical, geomechanical, and reservoir engineering data to characterize the in situ stresses, the natural fracture network and the controls on fracture permeability in geothermal reservoirs. INTRODUCTION The interaction of fracture flow systems with the present-day stress field is complex and accurate evaluation of reservoir behavior remains a significant challenge to the exploitation and management of fracture-dominated geothermal reservoirs. The objective of our approach is to develop an effective and efficient simulation to characterize, model and predict fractured reservoir performance during injection and depletion. If we can gain knowledge of the stress state, the distribution of natural fractures, and the properties of those fractures, then informed decisions can be made about favorable locations and directions to drill wells so as to maximize productive contact with the reservoir and provide sufficient heat exchange between the fluid circulating in the fracture system and the surrounding rock mass. Information about both in situ stress and fracture patterns from image logs is required as a minimum to predict optimal orientations to drill. The circulation of fluids and the contact area with the rock mass they sweep within the reservoir depends on characteristics of the fracture network involved, such as the intensity, size, orientation, volume and effective hydraulic aperture, of which limited measurements can be made directly within a well. Further, these characteristics change in response to reservoir stimulation or depletion. Thus, an understanding of the dynamic relationship between stress and fractures has to be applied to characterize the system implied by monitoring data acquired during hydraulic stimulation. The essential monitoring data includes: located microseismic data, pressure build-up and flow-rates in addition to standard wellbore fracture characterization. An integrated interpretation of the dynamic fracture system can then be applied to predict further stimulation or production behavior. To demonstrate this method we utilize a comprehensive data set from a basement granite reservoir in the Cooper Basin, the Habanero Field, for which both the present day in situ stress field and the natural fracture population are known (Fernández-Ibáñez et al., 2009). Key elements of this data set include modeled stress and pore pressure data, ultrasonic wellbore image data, microseismic data recorded during low flow rate long-term stimulation, and tracer tests.

2 CASE STUDY: HABANERO FIELD COOPER BASIN Overview Habanero is an engineered geothermal system (EGS) in the Cooper Basin in northeast South Australia operated by GeoDynamics Limited. The geothermal reservoir is a water-saturated, naturally fractured basement granite (250ºC at about 4300 meters) with permeability that was effectively enhanced by fracture stimulation. There is a sedimentary cover above approximately 3650m. Stress Regime Stress modeling results reveal high differential stress within the Habanero Field, where S Hmax is likely to be up to 150 MPa (Fernández-Ibáñez et al., 2009). The overburden, S v, which is significantly lower than S Hmax, is estimated by integrating the density recorded as a function of depth in the Habanero wells. Recently available numerical modeling results of wellbore breakouts indicate that the least horizontal stress, S hmin, is intermediate between S Hmax and S v. The mud weights used in the wells drilled in the region together with the drilling experiences reported in those wells provide valuable constraints on reservoir pore pressure. The pore pressure profile suggests a regional overpressured zone occurs below ~2,600m. The presence of coal seams in the stratigraphic sequence appears to be an additional source of local overpressures. High pore pressures occurring in the granitic basement are likely related to the occurrence of natural fractures. These stress and pore pressure analyses indicate that the field lies within a thrust faulting stress regime (S v < S hmin < S Hmax ). A 3-D structural grid of the Habanero Field was developed honoring structural and stratigraphic constraints, to provide a common framework for the geomechanical model and the other modeling and simulation analyses required for this study. This 3-D structural grid was then populated with the geomechanical properties. The 3-D overburden stress is derived using an analytical method which accounts for the effects of irregular surface topography and lateral density variations; effects not captured by standard vertical linear integration of the density (Holland, et al., 2010). The horizontal stresses are calculated within the 3-D structural grid based on the local pore pressure and overburden values in combination with modeled effective stress ratios. The resulting static 3-D geomechanical model contains all properties for direct grid-based extraction of the full set of geomechanical parameters required to calculate the mechanical coupling between the stress and Figure 1: Example of 3D GEM for Habanero showing static model of S Hmax. natural pre-existing fractures distributed within the reservoir. Fracture Characterization Using Wellbore and Microseismic Data A fracture conceptual model characterizes fractured bedrock in terms of fracture sets and orientation, spatial distribution of fractures, fracture intensity, fracture size distribution, effective hydraulic aperture, and mass balance aperture. Fracture intensity can be calculated based on plots of cumulative Terzaghi weighted fracture count (Terzaghi 1965). The spatial distribution of fractures may be correlated to geological or mechanical controls, such as structural features, lithology, beds, and structural curvature. Swarms of fractures usually appear as steep gradients on such plots and can indicate position of fracture zones distinct from the more sparsely fractured rock between those zones (Figure 2). Fracture sets and orientation are typically inferred from stereographic density plots interpreted from image log analysis (Figure 3a). Fracture size distribution cannot be determined directly from essentially 1-D measurements in a wellbore, and so typically modeling concepts such as power-law or log-normal distributions are used.

3 However, size distribution has a strong control on connectivity of a fracture network, the effects of which can be observed through flow-metering, well tests, interference tests or tracer tests. Hence, the fracture size parameters can be inferred from well test and production data. Hydraulic aperture is essentially inferred from the size of the pressure rise for a given injection rate. Mass balance aperture (fracture volume) controls the amount of fracture surface that is stimulated for a given injection volume and the breakthrough time for tracers. Fracture compliance affects the rate of change of pressure in the fracture system. Figure 2: Fracture spatial distribution and intensity can be interpreted from cumulative Terzaghi weighted fracture intensity profiles. A fracture conceptual model can be derived based on wellbore image analysis, well tests and tracer tests. However, using only these data requires several assumptions regarding fracture geometries and hydraulic properties that result in significant modeling uncertainties. In this study we interpret in a novel way available microseismic event data to more precisely define fracture geometry, and also use the microseismic data to calibrate the model for the dynamic behavior of the system during stimulation. a) b) Figure 3: Fracture sets and orientation distributions can be interpreted from stereonets based on (a) image data or microseismic data (b). Equal Angle Wulff net, lower hemisphere projection.

4 Figure 4: Located microseismic events during phased hydraulic stimulation colored and scaled by moment magnitude. An example of located microseismic data during the early stages of the Habanero-1 well stimulation is shown in Figure 4. By carefully animating the development of the microseismic cloud it is possible to infer fractures in the order in which they were stimulated. In this way, a realistic impression of the dynamic stimulation of pressure build-up and consequent fracture stimulation along connected pathways is obtained. Figure 5 shows effective hydraulic apertures inferred where slip events occurred for the main fracture intersecting Habanero- 1. Figure 6 shows this fracture along with some of the other fractures detected in this way in the vicinity of the Habanero wells. Although it is difficult to map an entire network using microseismic data, it is possible by defining the immediate network around the wellbore to use its properties as the basis for constraining fracture orientations (Figure 3b), size (Figure 5b) and connected fracture intensity. Further, the time at which fractures are stimulated gives an indication of hydraulic diffusivity and by inference hydraulic aperture distribution (see color symbols, Figure 5a), and the stimulated fracture area versus cumulative injected volume can be used to calculate a mass balance aperture (see Table 1). Fracture Geometry and Effective Hydraulic Apertures Inferred at Slip Events The initial stimulations at Habanero-1 were analyzed in this way for three different phases of hydraulic activity; 1&1B, 2&2B and 2C. Stimulation 3 was unsuccessful, and has not been used to provide aperture statistics. For each of these three stimulation phases, the total surface area of the stimulated network is calculated based on the interpreted fractures from microseismic data, and the average mass balance aperture can then be estimated from the injected volume of fluid, as shown in Table 1. Likewise, the slip events induced by each stimulation phase can be regarded as an interference test initiated with injection in the main fracture and the propagation of pressure indicated by the location and time of each slip event, thereby providing an estimate of hydraulic diffusivity (Streltsova, 1988). Effective transmissivities and hydraulic apertures can be estimated from the diffusivity by analogy to hydraulic tests performed in other granitic basements (e.g. Rhén et al. 2008). A summary of the geometric mean hydraulic apertures and transmissivities identified for stimulations 1&1B, 2&2B is presented in Table 2. The distribution of hydraulic apertures and transmissivities are shown in Figures 7a and b respectively. The suggested nominal ratio of about 10 between mass balance apertures and hydraulic aperture (cubic law aperture) is similar to that interpreted from the collation of extensive tracer tests in the Fennoscandian shield (Hjerne et al 2010), reflecting roughness attributes of the surfaces of the fractures and possibly additional volume associated with the immediately surrounding rock matrix.

5 a) b) Figure 5: a) Effective hydraulic apertures inferred where slip events occurred on a fracture intersecting Habanero 1 at an elevation of 4135m. b) Water conducting fracture size distribution interpreted from microseismic events. Fracture aperture scale is microns and radius scale is meters. Table 1: Example of inference of mass balance aperture from located microseismic events. Table 2: Stimulation Geometric mean effective apertures and transmissivities inferred from the microseismic data for each of the three phases of stimulation Number Of Stimulations Mean Hydraulic Aperture Mean Transmissivity 1&1B μm m 2 /s 2&2B μm m 2 /s 2C μm m 2 /s

6 Frequency [-] Frequency [-] Figure 6: Part of the inferred network of fractures local to Habanero 1 viewed from the southeast, and determined through analysis of microseismic events. Color indicates proximity of the fractures to frictional failure < D < < e < -7.0 Stim 1&1B Stim 2&2B Stim 2C -2.0 < D < < D < -1.4 Stim 1&1B Stim 2&2B Stim 2C -6.8 < e < < e < < D < < e < < D < < D < < D < < D < < D < 1.0 log 10 (Diffusivity) [m 2 /s] -5.6 < e < < e < < e < < e < < e < -3.8 log 10 (Aperture) [m] 1.2 < D < < e < < D < < e < -3.0 Figure 7: The distribution of effective hydraulic diffusivity (a) and apertures (b), estimated using microseismic data for stimulations 1&1B, 2&2B and 2C. Stress Aperture Coupling and Transient Flow Modeling Low flow rate injection tests were used to characterize the hydraulic properties of the fractures, their width, stiffness and strength properties that are often difficult to quantify, leading to the typically large uncertainties in predicted response to stimulation of fractured reservoirs. The initial, low flow rate injectivity test data recorded in the Habanero-1 well are shown in Figure 8a as pressure versus flow rate. The lower plot of Figure 8 shows the cumulative number of fractures that have slipped versus increasing pressure. As injection proceeds from an initial low rate to higher rate, welloriented fractures tend to slip; at higher pressures, less-well oriented fractures will also slip. Much of this hydro-shearing occurs at stimulation pressure less than the formation fracture gradient (the least horizontal principal stress, S hmin ). The shape of the measured pressure versus flow rate curve (symbols) is forward modeled (solid line) by varying the parameters of the relationship between equivalent aperture and effective normal stress and the frictional properties following Moos and Barton, (1)

7 a) b) Figure 8: Aperture-stress coupling calibration from Habanero-1 stimulations. a) Flow rate versus pressure for modeled (solid line) and measured (symbols) injection and number of fractures slipped with pressure (b). Table 3: Aperture-stress coupling parameters determined for the Habanero EGS field. Parameter Mean Standard Deviation Model Depth [m] 4,267 S Hmax [MPa] % S hmin [MPa] % S v [MPa] % P p [MPa] % S Hmax Azimuth [deg] 80 ±10 Cohesion [MPa] % Sliding Friction % a % A % A slip % B % B slip % where a is equivalent aperture, n the effective normal stress, a 0 A and a 0 A slip are the pre- and postslip apertures, B and B slip the pre-and post-slip 90% closure stresses. The aperture-stress coupling model parameters used to fit the data are provided in Table3. These fracture flow properties, stresses and the deterministic fracture data discussed above were used to explicitly model flow and transport through individual fractures which form the connected discrete fracture network (DFN). The modeling software, ConnectFlow, applied in this study is an efficient finite-element DFN method that uses a stochastic approach to generate networks of planar fractures that have the same statistical properties as those that are interpreted from field data. Fracture stress sensitivity is internally coupled to the flow simulation through the DFN with dynamic adjustment of aperture to effective normal and shear stresses (after Moos and Barton, 2008) and calibrated with microseismic data (positions and times of events) and injection data (rates and pressures). Figure 9 shows the stress-aperture coupled simulated and measured well pressures during the first four hours of Habanero-1stimulation. This plot corresponds to the pressure versus flow rate shown in Figure 8.

8 was performed over about 66 days with an average pump rate approximately 12kg/sec of brine (though there were some pump outages). 100 kg of the tracer 1,3,5-naphthalene trisulfonate was injected into Habanero-1. The water produced in Habanero-3 was re-injected in Habanero-1 to form a closed loop. Samples were taken from the produced water and analyzed. Figure 9: Measured and simulated well pressures during the first four hours of Habanero-1 stimulation. The result of modeled and measured stimulations 1&1B, 2&2B over about one month of phased stimulation are shown in Figure 10. In order achieve the level of match shown the main adjustments made were to fracture intensity and aperture-stress coupling parameters. In predictive mode, the simulations can be used to estimate likely pressure evolution, as well as extent, geometry and volume of hydro-shearing resulting from the stimulation (Figure 11). Having achieved a satisfactory calibration of the coupled fracture and geomechanical system based on monitoring data from the stimulation process the validity of the derived model is confirmed. To further verify the model we performed further testing by simulation of a post-stimulation circulation and tracer test. A dipole test with injection in Habanero-1 and production in Habanero-3 approximately 550m apart To simulate the tracer test a steady-state calculation of flow and transport through the DFN model was performed using the history of fracture slip calculated during the simulation of hydraulic stimulation. That is, fractures that were predicted to have slipped during stimulation were assigned a different aperture distribution and stiffness (see Table 3) than those that had not. Transport of solute was calculated using a particle tracking method through the network based on a flux weighted routing between pairs of intersections. This is a very efficient method to calculate a breakthrough curve for an inert tracer migrating with a steady-state dipole flow situation. Effects of re-injection of produced water can be accounted for by a convolution method. The fracture modeling approach used is necessarily stochastic as not all fractures in the system can be identified even with microseismic data. Hence, a Monte Carlo method of fracture generation was used. Generally, it was found that the simulation of the tracer test is more sensitive to DFN realization than the wellhead pressure during stimulation due to the sparse nature of the connected fracture network between injector and producer. For this reason, the tracer test was used to screen realizations that best mimicked concentrations and recovery characteristics at the producer. The properties of the fractures remained unchanged from one realization to the next. Figure 12 shows the simulated concentration in Habanero-3 for the best of 10 realizations attempted, and corresponding to the stimulation results shown in Figures 10 and 11. The simulations predicted a tracer recovery factor of 0.76, measurements gave 0.78.

9 injection rate [bpm] wellhead pressure [MPa] stimulation rate [-] C Measured Simulated /11 08/11 10/11 12/11 14/11 16/11 18/11 20/11 22/11 24/11 26/11 28/11 30/11 02/12 04/12 06/12 08/12 10/12 12/12 14/12 date Figure 10: Composite plot of simulation results against actual stimulation over about 1 month duration. The lower red curve shows the measured injected flow-rates. The light blue curve indicates the simulated pressure with the dark blue points showing the measured pressure (note the flat lines indicate periods when the gauge was no longer measuring well pressure). The green spikes indicate simulated rate of microseismic events with green dots indicating observations. Figure 11: Example simulation of slip events on fracture surfaces during a multi-phase stimulation. Part of the connected DFN network is shown around the stimulated well in the center (the distance between the two wells shown is about 500m). The stimulated fractures are shown as points colored by the time of slip; the geometry of each fracture is modeled using segments 20 meters on a side.

10 Figure 12: Comparison of tracer concentrations for post-stimulation circulation test between Habanero-1 and Habanero-3. The breakthrough curve is based on 50,000 particles tracks through the DFN model using the aperture distribution calculated from the simulation of the stimulation process. The re-circulation curve accounts for the re-injection of produced water. Geomechanical Sweet Spots for Stimulation Design Optimal well locations in this field are those which have the highest permeability to enable production and access to the resource, or in which such permeability can be created by stimulation. Typically, these are locations where there is the greatest likelihood of finding a dense network of preexisting fractures. Fracture flow is dominated by interconnected fractures in the reservoirs and stimulation is controlled by the intersecting fractures that are also connected to the wellbore. Using the DFN generated for Habanero we first determine fracture intersections and then calculate fracture connectivity; both the connectivity of the intersecting fractures (intersection length) and their connectivity to the wellbore. Figure 13a shows the resulting connected fracture network model for this field. As discussed, it is largely the stress-sensitive fracture sets that govern the drainage of fracture dominated reservoirs (Barton et al., 1995; Barton et al., 1998; Barton and Moos, 2009, Hennings, et al., 2012). Integrating the results from our 3D geomechanical grid with the DFN fracture distribution we can determine using Coulomb failure analysis the proximity to frictional failure of each fracture in the DFN. This provides a means to identify the optimal directions in which to drill wells to maximize productivity. As a measure of the stress sensitivity of each fracture we use the calculated shear stress,, divided by the critical shear stress, max. This ratio, referred to as ratio, lies between 0 and 1. Fractures with higher ratio values (~1) are considered stress sensitive; fractures with lower ratio values are considered stress insensitive. This measure is used to visualize the critical potential of each fracture surface and to model the potential production enhancement for a given stimulation plan. Figure 13b shows ratio, the proximity to frictional failure of the connected fracture network. By mapping the fracture intersections to the structural grid we can evaluate fracture sweet spots, reservoir zones with a probabilistically high fracture density (Figure 14a). We can also evaluate geomechanical sweet spots for stimulation design, zones with the highest density of stress sensitive fractures (i.e., ratio ~1, Figure 14b) SUMMARY The goal of discrete fracture network characterization is to define reservoir fracture architecture

11 stochastically using available deterministic data. In this study, we use wellbore, microseismic and pumping data to generate and calibrate a realistic inter-connected discrete fracture network for the Habanero field. A post-stimulation tracer test for dipole injection between two wells was then used to confirm the resulting model parameterization and to screen stochastic realizations showing solute transport characteristics most similar to those observed. The results highlight the importance of extracting optimum value from all available data, particularly microseismic and pumping data. They also reveal how sensitive the predictions are to the assumed fracture architecture. This highlights the importance of establishing deterministically the locations and connectivity of as many fractures in the network as is required both to characterize the local fracture system and to inform statistics of the wider undetermined stochastic system. Stochastic realizations conditioned on this local information are much more likely to provide meaningful predictions of system behaviour than those which are not.. a) b) Figure 13: a) Stochastic connected fracture network model. b) Analysis of the connected fracture network showing ratio a) b) Figure 14: a) DFN fracture intersections mapped to the Habanero structural grid and intersection of fractures with ratio > 0.6

12 REFERENCES Barton, C. A., M. D. Zoback, and D. Moos, 1995, Fluid flow along potentially active faults in crystalline rock: Geology, v. 23, no. 8, p Barton, C. A., S. Hickman, R. Morin, M. D. Zoback, and D. Benoit, 1998, Reservoir-scale fracture permeability in the Dixie Valley, Nevada, geothermal field: Proceedings of the Society of Petroleum Engineers/International Society for Rock Mechanics 47371, Eurock 98, Trondheim, Norway, July 8 10, p C. A. Barton, D. Moos, and K. Tezuka, Geomechanical wellbore imaging: Implications for reservoir fracture permeability, AAPG Bulletin, v. 93, no. 11 (November 2009), pp Streltsova T D, Well testing in heterogeneous formations. Exxon, Monograph. John Wiley and sons. Terzaghi R, Sources of error in joint surveys. Geotechnique 15(3): ACKNOWLEDGEMENTS The authors wish to thank GeoDynamics Limited for the use of their complete set of data from the Habanero Field and for their generous time assisting in its delivery and interpretation. We thank AMEC and Baker Hughes RDS for their permission to develop this workflow. We are also grateful for the helpful conversations and technical assistance of Marc Holland and Thomas Finkbeiner. Fernández-Ibáñez, D. Castillo, D. Wyborn, D. Hindle. Benefits of HT-Hostile Environments on Wellbore Stability: A Case Study from Geothermal Fields in Central Australia, Proceedings, Indonesian Petroleum Association, Thirty-Third Annual Convention & Exhibition, May 2009, IPA09-E-057. Hennings, P., P. Allwardt, P Paul,C. Zahm, R. Reid Jr., H. Alley, R. Kirschner, B. Lee, and E. Hough. Relationship between fractures, fault zones, stress, and reservoir productivity in the Suban gas field, Sumatra, Indonesia. AAPG Bulletin, v. 96, no. 4 (April 2012), pp Holland, M. M. Brudy, W. van der Zee, S. Perumalla and T. Finkbeiner, Value of 3D Geomechanical Modeling in Field Development - A new Approach using Geostatistics SPE/DGS Annual Technical Symposium and Exhibition held in Al- Khobar, Saudi Arabia, April Hjerne C, Nordqvist R and Harrström J, Compilation and analyses of results from cross hole tracer tests with conservative tracers. SKB R Svensk Kärnbränslehantering AB. Moos, D., and Barton, C. A. Modelling uncertainty in the permeability of stress-sensitive fractures. ARMA Rhén, I., Forsmark, T., Hartley, L., Jackson, P., Roberts, D., Swan, D., and Gylling, B. Hydrogeological conceptualisation and parameterisation, Site descriptive modelling SDM- Site Laxemar. SKB R-08-78, Svensk Kärnbränslehantering AB.