Quantitative Use of Time-Lapse Seismic to Update Reservoir Models (Summary of the the HUTS Project)

The History matching Using Time lapse Seismic (HUTS) project has developed techniques to use repeat seismic data along with production data to condition dynamic reservoir models during the history matching process. This European-funded project has finished and the expertise gained is now available to use on operational assets. Also, the software developed is commercially available (Eclipse HUTS - http://www.sis.slb.com/content/software/simulation/eclipse_pre_post/huts.asp ). HUTS is a collaboration between TotalFinaElf (now Total), Norsk Hydro, ENI and Schlumberger. The principal investigators were Olivier Gosselin and Romain Gonard, Total Sigurd Aanonsen and Ivar Aavatsmark, Norsk Hydro, Alberto Cominelli and Luciano Kovacic, ENI, Kieran Neylon, Schlumberger. Olivier Gosselin ( olivier.gosselin@total.com ) of the Total E&P UK PLC Geoscience Research Centre in London presents a summary of the project. Full details can be found in SPE 84464, presented at the 2003 SPE Annual Technical Conference and Exhibition.

Background

The HUTS project was set up in 2000 based on the strong belief that the emergent use of time-lapse or 4D seismic data should be quantitatively integrated into reservoir characterisation at the history matching stage, using this information as a new kind of dynamic observation in addition to well production history data to update reservoir models. Since the establishment of the Geoscience Research Centre (GRC), computer-aided history-matching (CAHM) had always been a main activity, focussed on gradient-based techniques, through the development of internal prototypes and external cooperation with Norsk Hydro and Schlumberger/Geoquest. A previous European project (1996-1997) led to the first commercial release of SimOpt in 1998. Then ENI-Agip, motivated to join in the development of both CAHM and 4D seismic, joined us in 1999 to prepare this multidisciplinary project, combining the skills of geophysicists and reservoir engineers, to develop expertise and to produce a new commercial version of SimOpt-Eclipse . The European contract officially started in December 2000 and lasted two years (with TotalFinaElf, Norsk Hydro, ENI, Schlumberger - John Barker was leader at that time).

When we started this project, some major oil companies had extensive practical experience in 4D monitoring, generally with a qualitative, pragmatic, but apparently successful, approach. However, details about the methods used were not published and were only available within internal software. N umerous presentations from contactors also showed fully integrated workflows, with nice virtual loops. Xuri Huang et al. [3] detailed a global workflow based on three levels of matching - seismic amplitudes, elastic impedances or fluid pressures and saturations, but no commercial tools were available. Generally speaking the qualitative use of 4D data was suggested, often assuming a fluid change interpretation, and sometimes proposing a global loop including forward seismic modelling. Examples of the successful use of 4D seismic were reported without resolving the incompatibility with production matched models. So this was precisely the challenge to be addressed.

Objectives

What 4D seismic - repeated seismic acquisition after field production - can bring to reservoir monitoring is now well known and proven. Reservoir monitoring might mean observation of the reservoir during production, to control the effectiveness of operations, or the use of this dynamic information to direct appropriate actions, but we believe that it should also involve using these observations to update the properties of the reservoir model. The project thus aimed to use the new information on fluid movements provided by repeat seismic acquisitions, jointly with production data, in an extended history-matching process to produce consistent models.

The main objectives were to:

• Focus on the reservoir engineering side, target the impact of 4D seismic to reservoir model updates.
• Ensure consistent matched models explaining both time-lapse seismic and well production data.
• Use established CAHM techniques, particularly gradient-based methods, and sensitivity analysis.
• Develop and implement an approach demonstrated on real cases, through internal prototypes.
• Develop a commercial tool available immediately after the end of the project, closely following partner specifications.

HUTS Approach

After brainstorming meetings with geophysicists, geologists and petro-acousticians, we opted for a matching loop at the elastic domain level, emphasising the importance of the petro-elastic modelling (PEM or rock physics module), which determines elastic properties of saturated porous media from simulated fluid and static rock properties, and makes the bridge between the fluid flow and the wave propagation domains (Figure 1).

Figure 1: Possible Levels of Matching

Fig 2: HUTS workflow

The main reasons for this choice were:

• It avoids the need to include within the iterative loop the cpu-time consuming forward seismic modelling; on the contrary each survey, separately or together, is inverted only once.
• It takes into account both the saturation and pressure effects (through PEM) and avoids a difficult and possibly inconsistent "inversion" in terms of fluid changes, because these pressure/saturation changes cannot be considered as observations independent of the reservoir model.

Our approach is thus a quantitative use of 4D seismic data, involving a simultaneous minimisation of the mismatch (objective function) between all types of measured and simulated data. The seismic contribution is defined in terms of elastic parameter variations within the reservoir - "observed" values obtained by inversion of the seismic signal, and "modelled" values, obtained by the fluid flow simulator coupled with a petro-elastic model. The main points are:

Global minimisation loop - Gradient-based technique to reduce both production and 4D mismatch, each term weighted by prior observation uncertainties (Figure 2).

Time-lapse seismic inversion - I mpedances (or whatever relevant parameters) mapped into the reservoir grid with associated uncertainty - posterior covariance matrix after inversion - to weight the seismic term (Figure 6).

Petro-elastic model (PEM) - Coupled with the fluid flow simulator ( Eclipse ), with gradient computations to be used by SimOpt ; the formulation is general - for future cases - and specific - for the cases studied in HUTS (Figure 5).

This approach was designed to produce one useful tool to assist the engineers in the global process, but was not intended to exclude qualitative and manual procedures. The cases studied have emphasised the need for combining both approaches.

The up-scale and down-scaling issues have been provisionally "solved" using only the reservoir grid scale, that means using seismic inversion which is able to provide better vertical resolution (like geostatistical inversion constrained by well log data and stratigraphic grid), horizontal upscaling of inverted impedances and including a correlation matrix in the objective function.

Project Work Packages

The work load was divided into seven main work packages, with specified leaders.

Prototype coding (Total) - HM4D , internal GRC code, shared with and further developed by the project partners. This prototype plays the role of SimOpt , driving the minimisation and calling Eclipse , matching pressure/saturation and elastic parameters as observations (both as absolute or changes between two surveys). It contains a PEM module.

Optimisation issues (Norsk Hydro) - The traditional least square formulation has been retained, with a standard optimisation algorithm. The balance between terms has been partially solved by focussing on the best possible covariance matrix error estimate. Determination and use of this seismic correlation matrix has been implemented. The gradzone analysis technique has been extended to the new 4D term.

Inversion techniques (Norsk Hydro) - The seismic inversion part was investigated separately for each case under study, with common lessons. 4D impedance cubes were produced using different methods (Figures 4 and 6). Exchanges of views were fruitful on the PEM, a general formulation has been developed bearing in mind both immediate and future applications. It has been implemented in HUTS-PEM software, and in a PEM-Eclipse module.

Demonstration cases (Norsk Hydro, Total and ENI) - To develop our approach we first applied it to a synthetic case. Using the PunqS3 case, we validated both pressure/saturation and impedance matching with a good synergy between production and seismic observations. We also obtained useful indications which appeared to be valid for the real cases (Oseberg - Alpha main part in ORELN2 formations, and Cervia and Amelia in the Adriatic Sea ). We obtained satisfactory results to validate the approach, but with no huge improvements in the reservoir models.

Commercial Software (Schlumberger) - The first year Schlumberger provided a new version of SimOpt handling pressure and saturation changes as new observations for history-matching. The HUTS PEM was implemented in Eclipse . A "pre-commercial" release of Eclipse and SimOpt was delivered in October 2002, handling elastic impedance and Poisson's ratio changes as new observations for history-matching. The official commercial 2003a release with 4D features is now available.

Reporting - Numerous reports sent to EC (management and technical, partly publishable) and many publications.

Figure 3: Objective Function (PunqS3)

Figure 4: Objective Function - Oseberg Case (Effect of Correlations)

Main Results

The principal outcomes of this two-year project are a demonstration of the feasibility of our approach , the acquisition of expertise , and the development of new commercial software tools.

Through both synthetic and real cases, the following points were demonstrated:

• P roof of concept on a synthetic case with confirmation that a model history-matched to production data could in fact be bad in terms of fluid distribution, illustrating the need to take both sets of information into account in a global process (Figure 2).
• For synthetic and real cases, and almost all regression runs, the synergy between minimisation of both types of observations was effective and the gradient-based regression approach successful in reducing both mismatches (Figure 3). HM4D and its PEM module proved to be efficient. New Eclipse and SimOpt versions are now available.
• The petro-elastic model (PEM) is a key element in achieving a successful outcome and this is obviously a challenge in real applications; the development of our PEM focussed on theoretical and general insight into the physics involved, taking best practice from the literature, internal experience and previous studies. The gradient computations, needed for the minimisation procedure, allow us to run sensitivity studies to almost any static and dynamic reservoir parameters.
• Work done with the Oseberg sector model emphasised the need for an initial static geological model consistent with the seismic base survey, not only in terms of structure but also in the petrophysical property distribution. The recommended approach is to first adjust the pore volumes using the absolute impedance values at time t=0 (base survey), with a loop around Eclipse initialisation followed by petro-elastic modelling; using gradient minimisation within this loop.
• The use of a correct estimate of uncertainties is crucial - quantification of the seismic inversion quality using multi-realisations or auto-correlation of a single realisation - to weight the seismic term in the objective function, to ensure better convergence (demonstrated on Oseberg case, Figure 4) to a better matched solution (demonstrated on PunqS3 case); a possible way was suggested (see References).
• To avoid the vertical downscaling issue, an inversion process which permits a reservoir grid vertical seismic resolution is preferred (constrained by well information and stratigraphic grid).
• Parameterisation is a key issue in history-matching and a successful match can only be obtained if pertinent reservoir parameters are chosen. The gradzone analysis technique (using first and second order derivatives) was successful for the PunqS3 case, but less easy to use for the real cases. The provisional conclusion is that it is still applicable but is most beneficial when associated with a qualitative analysis of all available data. This was the approach used for Oseberg and this is an area where further work is needed.

Figure 5: Simulated Acoustic Impedance Changes (1992-1988 first column, 1999-1988 second column)

Figure 6: Simulated vs. Inverted Impedance Changes (after 7 years)

Next Steps

The end of the HUTS project is actually a starting point for the project partners. We are now ready to apply our approach and tools to our operations and to interact with the operational assets, while improving our tools and consolidating our methodology :

• New workflow to build an initial static model consistent with the base-survey (3D seismic data).
• PEM formulation investigation for each specific new case, but also addressing the general up/down-scaling issues.
• Early impact of 4D and PEM across the 3D geo-modelling workflow.
• Scaling issues at the level of seismic observations.
• Alternative ways of getting uncertainties for the inverted impedance cubes in depth.
• Parameterisation strategies
• Regression strategies combining production and 4D data (sequentially which one first, etc.).
• New thoughts on objective function formulation and use of no-gradient optimisation.

References:

1. Gosselin, O., Cominelli, A., van den Berg, S., and Chowdhury, S.D.: "A Gradient-Based Approach for History matching of Both Production and 4D Seismic Data," Proc. 7 th European Conference on the Mathematics of Oil Recovery , Baveno, Italy, 5 - 8 Sept. 2000.
2. Gosselin, O., van den Berg, S., and Cominelli, A.: "Integrated History matching of Production and 4D Seismic Data," paper SPE 71599, presented at the SPE Annual Technical Conference and Exhibition, New Orleans , Louisiana , 30 Sept - 3 Oct. 2001 .
3. Huang, X., Meister, L., and Workman, R.: "Reservoir Characterization by Integration of Time-Lapse Seismic and Production Data," paper SPE 38695, presented at the SPE Annual Technical Conference and Exhibition, San Antonio , Texas , 5 - 8 October, 1997 .
4. Waggoner, J.R., Cominelli A., and Seymour, R. H.: "Improved Reservoir Modelling with Time-Lapse Seismic in a Gulf of Mexico Gas Condensate Reservoir ", paper SPE 77514, presented at the SPE Annual Technical Conference and Exhibition, San Antonio , Texas , 29 Sept - 2 Oct. 2002 .
5. Aanonsen, S.I., Cominelli, A., Gosselin, O., Aavatsmark, I, and Barkve, T.: "Integration of 4D Data in the History Match Loop by Investigating Scale Dependent Correlations in the Acoustic Impedance Cube," Proc. 8 th European Conference on the Mathematics of Oil Recovery, Freiberg, Germany, 3 - 6 Sept. 2002.
6. Aanonsen, S.I., Aavatsmark, I., Barkve, T., Cominelli, A., Gonard, R., Gosselin, O., Kolasinski, M. and Reme, H.: "Effect of Scale Dependent Data Correlations in an Integrated History Matching Loop Combining Production Data and 4D Seismic Data," paper SPE 79665, presented at the SPE Reservoir Simulation Symposium, Houston, Texas, 3 - 5 Feb. 2003.
7. O. Gosselin, S. I. Aanonsen, I. Aavatsmark, A. Cominelli, R. Gonard, M. Kolasinski, F. Ferdinandi, L. Kovacic, K. Neylon: " History matching Using Time-lapse Seismic (HUTS)", paper SPE 84464, SPE Annual Technical Conference and Exhibition, Denver , Colorado , 5 - 8 Oct. 2003.
8. "Eclipse Technical Description", 2003, and "SimOpt User Guide", 2003, Schlumberger GeoQuest .

Disclaimer: The material available on this website is designed to provide general information only. Whilst every effort has been made to ensure that the information provided is accurate, it does not constitute legal or other professional advice.
Please note: The Department of Trade and Industry cannot be held responsible for the contents of any pages referenced by an external link.