Reservoir simulation

Reservoir-Modeling Work Flow Introduces Structural Features Without Regridding

This paper proposes a novel work flow for structural-features modeling that allows the introduction of faults and other structural and nonstructural features to any simulation grid without modification.

jpt-2018-07-188575f1-hero.jpg
Fig. 1—(a) The seismic faults of a structural model. (b) The faults and related reservoir grid.

Current reservoir-modeling work flows are rigid, because modification to the understanding of the underlying structural model often requires a complete regeneration of the reservoir grid, which brings additional costs, delays, and incompatibilities with past calculations. This paper proposes a novel work flow for structural-features modeling that allows the introduction of faults and other structural and nonstructural features to any simulation grid without modification.

Introduction

The modeling of hydrocarbon reservoirs is a multiscale and multidisciplinary process that usually involves several months for advancement from seismic interpretation to reservoir simulation. One crucial moment in the life of a reservoir model is the creation of its reservoir-simulation grid. This involves the encoding of the so-called structural model (Fig. 1 above), a set of surfaces representing interfaces between different geological features. A given grid cell can belong only to one side of each of these surfaces, generating geometrical tension on the grid. Once the reservoir simulation grid is created, every aspect of the simulation depends on it.

Several weeks or even months are spent building and quality checking the reservoir-simulation model, which is completely based on the cell division of the space determined by the reservoir grid. Simulations are then run and compared with actual data, when it is available, or used for conceptual design of production methods and structures. Other times, the model already exists and new production data are acquired. These simulations and comparisons sometimes reveal fundamental flaws in the simulation model, such as the existence of an unseen or underestimated fault that was not included or the need to model differently the flow across a fault because a fault relay was ignored. In such cases, the engineer can decide either to ignore the issue, leaving an inaccurate model in place for the rest of the field’s life but retaining compatibility with previous simulations, or to modify the reservoir properties manually to mimic the missing feature. These manual modifications, however, are extremely costly, and the result is often nongeological and possibly inconsistent with future evolutions of the model.

A final option is for the seismic-­interpretation team to add the missing fault to the structural model. This may require many steps, ultimately changing the geometry and number of cells. Expected timelines will be delayed, the budget will grow, and some development decisions will have to wait for the new simulations. Because of this, in future projects, the engineer will take great care to ensure that every trace of a fault is included in the first version of the structural model. That way, transmissibility multipliers can always be set to unity and the faults can be removed de facto from the actual simulation if need be. This strategy, however, creates large, unwieldy models, often with more than 200 faults.

This paper presents a new tool capable of introducing surfaces implicitly in a reservoir grid. The surface could be picked from seismic data or synthetically drawn, and it could represent any geological feature such as a fault, an unconformity, or the external envelope of a body. This tool allows for new, fast work flows in reservoir modeling and an unprecedented flexibility in structural modeling.

New Work Flow

The objective of the new tool is to communicate directly with the simulator by modifying the simulation deck so that the new surface is taken into account as a flow modifier during simulation. The surface is represented in the grid as a set of intersected cell-to-cell connections. The representation language can be understood by most simulators and allows easy adjustment of transmissibility multipliers for the newly introduced feature, as a global value on the whole surface or as a connection-dependent function.

The intersecting algorithm consists of an alternating propagation process that follows both the topology of the surface, represented as a triangulated surface, and the topology of the reservoir grid. This process requires a first step that consists of seeding, during which the first cell is found to start the search, with a complexity of O(n log n), followed by an alternating propagation over the grid and the triangulated surface, with complexity O(k), where n is the number of cells in the grid and k is the number of cell couples that represent the surface in the grid. With real-world data, the computational time is on the order of a couple of hours for a couple of hundred surfaces intersected with a faulted reservoir grid of 20 million cells.

The output of this algorithm is a set of cell couples arranged such that their centers rest on different sides of the introduced surface. The fault side on which every cell lies is also determined. This leads naturally to a grid property coloring the cells on opposite sides of the original surface differently. Also, a new triangulated surface is built, with the polygons resulting from intersecting the opposed faces of each cell couple. Such a surface is always water-tight and, most importantly, it represents exactly the cell-to-cell connections in the reservoir grid that will be affected by the newly introduced feature.

The tool also outputs files that are formatted to be read by most common simulators. These files allow, by means of a simple text editor or a dedicated tool also already available, the setting of transmissibility multipliers on the connection sets of each of the features introduced. Transmissibilities on the intersection between connection sets of different features (cell-to-cell connections intersected by several surfaces) are singled out automatically and can be handled with simple rules such as “most restrictive,” “less restrictive,” or “harmonic average.”  

Use Cases

The paper presents two basic use cases of the tool: introducing an underestimated fault and splitting a fault, both cases in a reservoir-simulation model that had been in use for some time. The model contains more than 150 faults, and its reservoir grid has more than 18 million active cells. A complete rebuild of the model would take months of work by a dedicated team, whereas the new work flow required only a couple of days.  

In the underestimated-fault case, a fault was observed in the seismic data that was not deemed important to the simulation because of its very small throw. Consequently, it was not included in the structural model used to build the reservoir grid. However, during history matching, correctly matching the pressure profile for a well going directly through the fault proved to be impossible. Including the fault in the grid was not feasible because that would have required a costly regridding and loss of any possibility of comparison with older simulations. Using the proposed work flow, the fault was included implicitly, leading to an improved pressure matching on the problematic well without affecting the matching in other wells. Because the reservoir grid was not modified, the whole operation required only a few hours.

In the split-fault case, two faults were simplified into one in the original structural model and a single transmissibility was assigned to them. As a result, during history matching, no single transmissibility multiplier was able to match historic and simulated pressure for a well situated in the vicinity of the fault. In order to obtain a more-consistent geological profile and simulate the relay missing from the structural model between the faults, the fault surface was split and the work flow was run. The merged fault was implicitly replaced by the two newly created faults in the model without the need for regridding. A subsequent history match was able to match the pressure history at the well better without affecting the matching for all the other wells. In the matched simulation, transmissibilities for the two faults differed by two orders of magnitude, indicating that the role played by the relay was significant.

Conclusions

The tool presented here finds the set of cell-to-cell connections in a reservoir-simulation grid intersected by a triangulated surface. For large projects, this computation is performed in a few hours, regardless of the type of reservoir grid. All that is required from the reservoir grid is an application programming interface to navigate its cell-to-cell topology. The triangulated surface could be exported by any 3D drawing software, as long as it is in a standard format.

This tool also allows the introduction of faults, horizons, channels, salt bodies, or any other geological feature as an interface in the reservoir-simulation model and allows easy control of the transmissibility through them. The tool would also accept nongeological surfaces as input for the computations, opening the door to fast conceptual designs of volume partitioning or testing the existence of features invisible in the seismic but visible in production data. This freedom enables many different work flows that were not realistic before.

This article, written by Special Publications Editor Adam Wilson, contains highlights of paper SPE 188575, “A New Shortcut Work Flow in Flexible Reservoir Modeling: Introducing Structural Features Without Regridding,” by Alejandro Rodríguez Martínez and Stefano Frambati, Total, prepared for the 2017 Abu Dhabi International Petroleum Exhibition and Conference, Abu Dhabi, 13–16 November. The paper has not been peer reviewed.