Enhanced recovery

Analysis of Steam/Solvent-Coinjection Processes by Use of Dynamic Gridding

Hybrid steam/solvent processes have gained importance as a thermal-recovery process for heavy oils in recent years.

jpt-2015-03-fig1gridding.jpg
Fig. 1—Temperature and pressure horizontal profiles for a fine steam/solvent-coinjection model with different levels of activity in some control properties of dynamic gridding. OMF=oil molar fraction. GMF=gas molar fraction.

Hybrid steam/solvent processes have gained importance as a thermal-recovery process for heavy oils in recent years. Among the identified physical mechanisms that play a role during these processes are heat-transfer phenomena, gravity drainage and viscous flow, solvent mass transfer, and mass-diffusion/-dispersion phenomena. In this paper, a study of sensitivity to grid size is described. Ideally, this work will provide some insight into methodological aspects to be considered when hybrid steam/solvent processes are modeled.

Introduction

Recent studies concerning the size of the liquid-solvent-rich zone where molecular diffusion and dispersion occur have implied that a detailed representation of the solvent/steam-chamber edge is necessary in the numerical model. To represent this front appropriately, the authors propose application of fine-grid models to represent the steam/solvent condensation zone and, in parallel, activation of the adaptive-mesh refinement (AMR) option for amalgamating the internal and external portions of the steam chamber zone. It is expected that, with this simulation strategy, a detailed representation of the front will be caught by a fine model, while a coarse model will be sufficient to represent the heat-transfer phenomenon that dominates in the noncritical zone. For a discussion of the amalgamation process, please see the complete paper.

Model Description

Synthetic Reservoir Model. A synthetic reservoir representing a generic formation of Athabasca oil sands (western Canada) is used in this work. The reservoir consists of a pay zone rich in oil. The pay zone is 45.5 m in length (x-direction) and 50 m in thickness (z-direction).

For simulating the steam-assisted-gravity-drainage (SAGD) recovery process, a horizontal production well with a length of 850 m (y-direction) is placed 1.5 m above the bottom of the pay zone. A horizontal injection well with the same length is situated parallel to the producer well with a 5-m vertical well spacing. The horizontal spacing between well pairs is approximately 91 m. Because of symmetry and the assumption of a homogeneous system, the authors modeled only one-half of the unit. A nonuniform 2D Cartesian grid system is used in the simulation. This grid contains five gridblocks in the x-direction of 0.1 m and then 90 gridblocks of 0.5 m. The number of gridblocks in the z-direction is 50. A fraction equal to 0.5 is specified for both wells in order to adjust the corresponding well indices to the one-half representation of the reservoir.

Initial oil and water saturation in the pay zone are 0.8 and 0.2, respectively. Initial reservoir temperature is 10°C. The reservoir formation consists of clean sand. The absolute permeability in the pay zone is 2 darcies in the horizontal direction while vertical permeability is three-fifths of the horizontal permeability. Porosity is 35%. Heat losses from and to overburden and underburden are specified. A live-oil fluid model is used in the simulation. Water, heavy oil, methane, and solvents (n-hexane or n-butane) are specified in the numerical model. The molar fraction of methane dissolved in oil is 0.037 at reservoir conditions. Phase-partition coefficients for the K‑value function were taken from the literature for the solvents. Heavy-oil viscosity is 8,028 cp at 50°C. A decreased function of oil viscosity as a function of temperature was used for a heavy Athabasca oil. Oil-phase viscosity was obtained by the natural logarithmic mixing rule currently used in the simulator. Values at 150°C are interpolated from an adjusted power-law function for the solvents. Gas viscosity of hydrocarbon components is fixed at .01 cp. No solvent-dependency effect in the relative permeability endpoints was considered in the simulations. Diffusion and dispersion coefficients were not specified in these simulations because there were not enough data in the literature related to the scaled solvent-dispersion coefficients. An asphaltene-precipitation model and water dissolution in oil phase were not considered in this model. It is expected that the current model simplification permits understanding of the main phenomena that occur during the steam/solvent-coinjection process without specifying dispersive processes explicitly.

Operational Conditions. The steam injector is controlled in pressure and in steam-injection rate. The injector well is operated at a maximum bottomhole pressure of 25 (or 35) bar and maximum water injection of 450 m3/d for the SAGD pair studied in this work. The steam quality is 90%. A preheating period of 3 months is performed before the steam injector is opened. The steam/solvent-coinjection process is analyzed for only 5 years to minimize the border effect in the numerical results. The operational strategy of this study is as follows unless stated differently: 1 year of steam injection, followed by 1 year of steam/solvent coinjection and 3 additional years of steam. Maximum water injection is not reduced explicitly during the coinjection period. Solvent/steam volume ratio is 0.20 volume fraction unless otherwise noted.

Steam/Solvent-Coinjection Process in a Fine Model vs. Fine Model With Dynamic Gridding

The main purpose of this activity is to select the appropriate parameters that can give results similar to those obtained from a fine model when a hybrid steam/solvent process is run. A steam/n-hexane-coinjection model is built to compare results obtained from a fine model with an elemental gridblock of 0.1×0.1 m and from the same fine model running with the dynamic-gridding option. A period of 1 year of steam/solvent coinjection was studied in this section because of high calculation time for the fine model. The amalgamated zones are defined as 1×1 m, so a set of fine cells can be lumped into a big one 10 times as large.

The resulting coarse cell has a surface 100 times greater than that of the elemental fine gridblock. The amalgamation process occurs when calculated property gradients are smaller than property gradients specified in the model. If changes in the selected properties exceed the defined threshold, a fine gridding is kept for those cells.

For running the AMR procedure, a sensitivity analysis is performed for three selected properties that were identified as appropriate to follow the steam/solvent front. These properties are temperature, oil-phase mole fraction, and gas-phase mole fraction. In general, temperature and oil-mole-fraction changes define the location where steam and solvent begin to condense. The gradient in gas mole fraction is also selected to indicate the point where condensable components change from the gas phase to the liquid phase. The maximum gradients in temperature, oil molar fraction, and gas molar fraction are defined as 10°C, 0.05, and 0.05, respectively. Five timesteps are taken as the minimum interval to perform checks to evaluate changes caused by dynamic amalgamation. The following four cases are tested to evaluate the effect of the trigger properties and selected gradients in a fine steam/solvent-coinjection model:

  • Three properties
    • Temperature (10°C),
    • oil molar fraction (0.05), and
    • gas molar fraction (0.05)
  • Two properties
    • Temperature (10°C) and
    • oil molar fraction (0.05)
  • Only one property
    • Temperature (10°C) or
    • oil molar fraction (0.05)

Fig. 1 above, Fig. 2, Fig. 3, and Fig. 4 show some horizontal profiles in a selected reference line (8 m from the reservoir bottom) for the studied cases. All cases injected steam and solvent for 1 year by use of a fine model (0.1×0.1 m). Injected solvent is n-hexane, while solvent/steam volume ratio is 0.10 volume fraction. A reference case was solved in a fine model without dynamic gridding. At the reference line, pressure is constant and temperature is equal to saturation temperature at steam partial pressure in the gas mixture. At the center of the steam chamber, solvent presence in gas phase is relatively low, to alter the steam-chamber temperature significantly. This results in local temperature close to steam-saturation temperature. At the steam-chamber edge (8 m from the steam-chamber center), steam begins to condense when local temperature equals the corresponding temperature of steam partial pressure. Solvent also condenses because its saturation temperature is close to that of steam at injection pressure. Water and oil saturation increase because of simultaneous water and solvent condensation. Condensed solvent is counted in oil saturation. The increment in solvent mole fraction in the oil phase creates a liquid-solvent-rich zone. This high solvent concentration reduces oil viscosity significantly, which consequently improves oil mobility. A priori use of two or more properties to control the dynamic-gridding process gives similar results. However, calculation time is significantly different when multiple properties are used to control the amalgamation process. Calculation time for the fine model was 106 hours. The use of three properties spends approximately twice as much time as use of two parameters. Results suggest that the use of dynamic gridding with two parameters can generate a 15-fold increase in speed when compared with a fine model. Profile results and calculation time show that dynamic gridding with two parameters is appropriate for mimicking results obtained with a fine model applied to the steam/solvent-coinjection process.

jpt-2015-03-fig2gridding.jpg
Fig. 2—Saturation horizontal profiles for a fine steam/solvent-coinjection model with different levels of activity in some control properties of dynamic gridding.

 

jpt-2015-03-fig3gridding.jpg
Fig. 3—Oil mole fraction horizontal profiles for a fine steam/solvent-coinjection model with different levels of activity in some control properties of dynamic gridding.

 

jpt-2015-03-fig4gridding.jpg
Fig. 4—Oil-viscosity and oil-mobility horizontal profiles for a fine steam/solvent-coinjection model with different levels of activity in some control properties of dynamic gridding.

 

Condensed solvent in oil phase expands as a function of time during 1 year of steam/solvent coinjection in a fine model without dynamic gridding. The solvent-rich zone enlarges with time because global solvent concentration increases in the system during the injection process. Results suggest that the selection of a decimetric scale to model the liquid-solvent-rich zone is necessary to catch physical phenomena that occur at the steam/solvent-chamber edge.

The identification of properties and definition of thresholds that define the water/oil interface by use of dynamic gridding were performed for three steam/solvent-coinjection cases: n-hexane/steam, n-butane/steam, and propane/n-butane/n-hexane/steam. The use of property gradients that trigger the amalgamation option for all solvents evaluated resulted in good consistency with those obtained from the fine model.

Effect of Gridblock Size

Once parameters that control the amalgamation process have been defined, the minimum size needed for modeling the steam/solvent-coinjection process at field scale is determined. The main objective of this part is to define the elemental gridblock size that permits capturing the steam/solvent liquid zone. The authors evaluated the following elemental gridblock sizes:

  • 0.2×0.2 m
  • 0.5×0.5 m
  • 1×1 m
  • 2×1 m

A steam/solvent-coinjection process was run for 5 years, starting with 1 year of steam injection followed by 1 year of steam/solvent coinjection and then 3 years of steam injection. Relatively large gridblocks (2×1 m) were selected to show the effect of the coarse representation in a field model. Considering that calculation time for the 0.5×0.5‑m case was 18 times faster than that of the 0.2×0.2-m case, we conclude that performing simulations with Cartesian gridblock sizes of 0.5 m is adequate to obtain a good resolution of the phenomena occurring at the front edge.
For a discussion of the effect of, and performance comparison of, three different solvent types (pure n-hexane, pure n-butane, and a mixture 50 vol% n-hexane and 50 vol% n-butane), please see the complete paper.

Effect of Injection Pressure

The effect of injection pressure was tested for different fluids. Injection pressure was fixed at 35 bar during a 5-year injection period. Injecting saturated steam at higher pressure implies higher saturation temperature. Oil-mobility profiles are surprisingly high for all solvents when steam/solvent is coinjected at high pressure. This uniformly high oil mobility results in a better bitumen-recovery factor for n‑butane and n-butane/n-hexane cases with respect to values obtained at low pressure. When n-butane is injected at high pressure, oil recovery is similar to that obtained for n-hexane. This confirms that n-butane performs better at high injection pressure (35 bar) than at low injection pressure (25 bar).

Conclusion

The use of coarse models causes a reduction in the maximum of solvent concentration in the oil phase, which directly affects the front displacement. The calculated local temperature in a coarse model also affects the location where steam and solvent condense because of insufficient discretized temperature and pressure. This results in a rough front representation, which results in a less-efficient thermal-recovery process.

This article, written by JPT Technology Editor Chris Carpenter, contains highlights of paper SPE 169075, “A Methodological Analysis of the Mechanisms Associated With Steam/Solvent-Coinjection Processes by Use of Dynamic Gridding,” by A. Perez-Perez, M. Mujica, and I. Bogdanov, CHLOE, and J. Hy-Billiot, Total, prepared for the 2014 SPE Improved Oil Recovery Symposium, Tulsa, 12–16 April. The paper has not been peer reviewed.