In production since 1982, the Valhall reservoir in the North Sea has recently entered a new phase of enhanced recovery with the commencement of a pilot waterflood to maintain pressure and enhance oil recovery. Valhall has long been considered a fractured reservoir with fractures ranging from small cracks to seismic-scale lineaments, resulting in a high level of uncertainty as to how these fractures would behave under waterflood.

To help understand the nature of the reservoir’s fractures and reduce uncertainty about the specific controls and location of the main productivity conduits, a phased program of data analysis and modeling was undertaken.

Picking the right model

The most common method for modeling fractured reservoirs is through conventional continuum simulators. However, these models approximate the behavior of fractures, and it was thought that they would fail to capture some key aspects of the reservoir.

Figure 1. Location map of the Valhall field. (Figures courtesy of Golder Associates)
Discrete Fracture Network (DFN) models held more promise at the scale of the planned waterflood. DFN models represent the fractures as discrete, largely planar objects with defined geometric and dynamic properties. As each fracture has specified fluid flow properties — permeability, compressibility and aperture — this means the DFN approach allows the integration of both static and dynamic data together in a manner that is far more representative than with a continuum approach.

Like any model, the DFN approach is a simplification of reality. Trying to represent all the fractures would make the model unwieldy and slow, so the model is set so it captures just the most permeable part of the fracture network — about 10% of the geologically identified fractures.

The initial goal was to develop a conceptual fracture model of how the various fracture populations influenced fluid movement within the reservoir. The first step was to study the data for each well, including well logs, drilling history, production history and well tests, to find the most likely conceptual model and to eliminate hypotheses that did not fit that data.

Figure 2. DFN model of the seismic fault network showing the classified faults, “red,” “green” and “yellow,” based upon their permeability, orientation and connectivity.
With these well data, a range of possible conceptual models was identified within the reservoir. It was important to synthesize several issues within the conceptual model, including generalized flow patterns laterally and vertically through the reservoir. There are three main feature types within the Valhall Chalk: through the chalk matrix pores; smaller fractures including joints, bedding plane features, hard grounds and sub-seismic faults; and the seismic-scale faults.

The analysis of the well data, well locations and fault proximities identified some generalized flow styles and also helped identify the most likely conceptual fracture model. This model indicated that seismic faults appeared to be laterally well-connected and provided the dominant flow pathways through the reservoir. It also indicated that sub-seismic faults and other fractures are not significantly permeable relative to the major conductors and could be considered homogeneous at the scale of the well tests.

Building the DFN model

The simplicity of the conceptual model meant that its implementation within a DFN environment was relatively straightforward, involving taking the seismic fault map from the pilot area and converting it to a DFN model. Some editing of the fault network was needed as a gas cap obscures much of the southern part of the pilot area, and many of the faults in Valhall have limited offsets. The editing involved extending faults to abut and terminate against other faults, where this appeared reasonable, to compensate for the lack of seismic resolution near the faults’ tips. Faults were classified into groups according to their azimuth, abutting relationships and the interpretation of whether or not they may have enhanced permeability.

The key test of a successful DFN model is whether it can replicate the dynamic response as seen at the wells. To do this, a detailed review of the available pressure transient data was undertaken.

Pressure transient synthesis

Well tests are arguably the best data set for investigating the geometry and connectivity of the fracture system as it extends away from the well bore. The current well test analysis focused on using the pressure derivative curve from available pressure transient data as an indicator of the nature of the flow geometry as observed by the well test, useful for building the conceptual model and for conditioning the DFN models.

Widespread testing in the pilot area of the Valhall reservoir has provided about 80 pressure transient tests. Plotting all the pressure derivative curves in a rate-normalized manner allowed the display of all the test data in a common space to identify spatial trends in transmissivity and flow geometry.

Figure 3. Snapshots of pressure drawdown resulting from well test simulation at 10, 100, 1,000 and 10,000 hours. (a) At 10 hours flow is dominated by the fault, which intersects the borehole and the segment of the motorway nearest to the well. (b) At 100 hours the drawdown has clearly spread into the matrix as well as the fault network. (c) At 1,000 hours the pressure drawdown “signal” has reached the end of the motorway and is spreading into the broader fault network. (d) At 10,000 hours the pressure drawdown has reached the boundaries of the model.
Grouping the well test curves into common styles involves examining the character of the derivative response as well as a subjective examination of each well’s geological setting. The different well tests were grouped into one of four classifications based upon the location and test response:
• Group 1. Wells on or near the high permeability “motorway,” with well tests showing high permeability and a flat or increasing late-time derivate.
• Group 2. Wells close to “red” trend faults, with lower permeability and a slower increase in the derivative than Group 1.
• Group 3. Near “yellow” faults, with a wider spread in permeability and a range of derivative shapes at late time.
• Group 4. Near “green” faults, with an increasing derivative slope of about 0.5.
The DFN model was then used to simulate flow in representative wells in each group to see if the chosen conceptual fracture could produce the right dynamic response and to determine the dynamic parameters for the fault network.

Testing the conceptual model

Given that our objective was to see if the model would reduce uncertainty about waterflood, it was enough to show that “generic” well test responses can be matched by using reasonable properties while maintaining a common conceptual model.

The DFN models were built in Golder Associates’ FracMan software, which also allowed the simulation of pressure transients through the model. By modifying the hydraulic properties of the various fault elements and the effective matrix properties, the simulated pressure derivative curves could be matched to the real test data. Well tests representative of the four different groupings were simulated using basically the same model with only minor modifications, giving confidence that the connected fault model was the most likely conceptual fracture model.

Conclusions

The DFN testing of the conceptual model gave us a better understanding of the distribution of permeable faults and the effective permeability values for specific fault trends. The calibrated fault model was subsequently imported to a conventional simulator to run a range of sensitivities which showed that the location of the seismic faults was a critical factor for predicting waterflood. Consequently, the waterflood was carried out in a phased approach and included a significant seismic and well-based monitoring program.

The simulation of key well tests for each of the four well categories showed that the same model could reasonably explain a wide range of behaviors, increasing our confidence in the conceptual fracture model. Being able to integrate complex structure and dynamic flow information within the DFN environment provided a powerful tool for building a consistent reservoir model. This work confirmed the view that Valhall should be considered a faulted rather than a classic fractured reservoir, and this changed the approach to the Valhall partnership’s waterflood plans.