SPE 39754

W
e
D . .

Society of Petroleum Engineefe

Survival of the Fittest an Optimised Well Location Algorithm for Reservoir Simulation
G. Santellani, AGIP S.p.A., B. Hansen, Norsk Agip, and T. Herring, Norsk Agip

Copyrfght 1998, Socieiy of Petroleum Engin-rs, ~ls pawr Integrated 1e98. was prepared for presenta!lon Mcdelling for Asset Management

Inc.

at the 1998 SPE Asia Pacific Conference on held In Kuela Lumpur, Malaysia, 23-24 March

geological descriptions and development scenarios have to be verified in short time, such as geostatistical realizations. The technical advantages of such a process are: . Consistent methodologies allow for true comparison between models. . Added value and reserves when compared to the traditional manual procedure. . Significant time saving. Introduction Currently, the use of a 3D reservoir simulator is recognized as one of the primary tool in defining the optimum production strategy in complex fields. Simulation is the only way to describe quantitatively the flow of multiple phases in heterogeneous reservoirs. The 3D simulator proves to be particularly important when it comes to the planning of number of production wells and their locations. Unfortunately, this process normally means a tedious and costly trial and error process, where the final result depends on the ability of the reservoir engineer to fully understand the reservoir behavior and the operational Iimits. This paper presents an alternative approach to the manual optimization of the production well locations in a 3D simulator model. This approach is automated and combines the simulation model with an external optimization code. The code follows a procedure similar to the natural selection and applies the principle of survival of the fittest to all the vertical production wells that is possible to generate in the simulation grid. The selection process works in steps. At the end of every step, a fitness function is defined for each production well based on results of a production forecasts. Automatic well location routine The optimization program has been built in connection with the in house 3D simulation model, however the code is general in nature and could be adapted to any commercial simulator changing the reading format of the data. The routine is a combination of FORTRAN codes and simulation runs. A UNIX script controls and execute the FORTRAN codes and simulation runs. The optimization 255 process is divided in three phases:

This pa~r wss se!acfed For presemtlon by an SPE Program Committee folloting review of Information contelnad In an abstract submitted by the author(s), Contents of the paper, as presented, have not been reviewd by the Smlefy of Patroleum Enginaers and are subject to mnwcffon by the author(s). The material, as presented, does not necessarily reflect any ~ition of the Society of Petroleum Engineers, its officers, or members. Pawrs presemed at SPE meelings are sub~cf to publication review by Editorial Committees of the Society of Petroleum Engineers, Electronic reproduction, distribution, or storage of any part of this paper for commercial purposes wfthout the written consent of the Society of Petroleum Engineers is pmhiblted. Permission to rapmduce In prfnt is restricted to an abstract of not more than 300 wrds; illustrations may not fm copied. me abstract must contain conspicuous a~dedgment of Were and by whom the pa~r wss presented. Wr~te librarian, SPE, P.O. Box 8X836, Rchardacn, ~ ~83-3836, U. S. A., fax 01-972-952-9435.

Abstract Traditionally, the optimization of well locations using a numerical simulator is time demanding and based on a manual trial end error process, This paper develops an automated technique to locate production wells. The automatic process is based on an algorithm that combines a 3-D simulation model with an external optimization code. The process is genetic in nature. Starting from a maximum well count, it proceeds in steps selecting a set of wells at the end of each 3-D simulation forecast and stops when a desired number of produci ng wells is reached. Two applications of the automatic well location process are presented. The first example considers the Ekofisk field, the largest oil field in the Norwegian sector of the North sea, with more than 20 years of production, water and gas injection. In this case, 32 new well locations have been automatically identified. The second example applies the process to the Sm@rbukk field, a very complex sandstone reservoir with oil and gas condensate layers and dry gas reinfection. 18 well locations have been automatically defined and connected to 6 subsea templates. In both cases, it has been possible to manage specific topics related to the field and to obtain a large increase in the economic value, accelerating the production and increasing the final field recovery significantly. The total process run time does not exceed 24 hours. This new methodology proved to be reliable and quick and is strongly recommended in studies where complex simulation work is required to optimize well locations and when different

G. SANTELLANI,

B. HANSEN AND T. HERRING

SPE39754

Phase I, wellgeneration. During this phase thecode makesa ~d blocks selection according to given constraints, based on and potential production petrophysical properties performances. It then generates all the possible vertical production wells connecting the selected grid blocks. AII the information needed at this stage are read, for each active block of the grid, either from the 3D simulation output files at the end of the history match or, if the field does not have any production history, from the output files at the end of the initialization process. Several parameters, like the fluid saturations, or more complex functions, for instance the production potential, the water cut, the gas oil ratio or the gas water ratio, can be selected by the user to determine if a block should to be open to production. At the end of this first phase an input file for the simulation model is built and a grid map displaying the new well locations is generated. No new wells are completed in blocks already containing existing wells to avoid calculation problems.. At the end of phase I, the user can optionally decide to start directly the third phase of the process or to enter the second phase. Phase II, static wel[ screening. Optionally, a selection based on well properties could be introduced to reduce the number of vertical wells to be processed during the third phase. For instance, the well selection could be based on a minimum number of layers open to production or the well production potential. The main objective of this phase is to reduce CPU time and computer memory required during Phase III. At the end of this phase a new input file for the simulation model is built and a grid map displaying the well locations is generated. Phase III, dynamic optimization. This third phase is the real optimization phase where the final selection is made. A simulation run is submitted with the production well locations resulting from the previous phases. At the end of the production forecast the wells are ranked according to a ~tness finction and a new simulation input file is written deleting a defined number of wells according to the well ranking then a new run is submitted. The iteration process stops when a pre-defined number of wells is reached. During this phase the process is completely automatic without any manual intervention once it starts. The ~tness function has to be defined by the reservoir engineer according to the final purpose of the reservoir study. It can simply be function of the well final hydrocarbon recoveries or function of the discounted well production rates. Penalties related to excessive water or gas production can be add in the function to match the field production constraints. In order to assure an acceptable two subroutines verify that: areal distribution of the wells

if in the list of wells that have to be deleted there are more than one well within a given interference radius r the ~ttest is kept for the next run. q If a well, in the list of wells that have to be deleted, does not have other production wells within a maximum well distance R it has to be kept for the next run The Fust condition avoids deleting a well when its production is reduced due to nearby wells and allows a double check in poor production areas. The second conditions guarantee a minimum well distribution. Both conditions could be turned off assuming the radii r and R smaller than the minimum grid block dimension. All wells are open , during the production forecast, at the same time to allow a fair selection. Ekofisk field - reservoir description The Ekofisk field was discovered in late 1969 . During its lifetime the field has undergone various phase developments from initial natural depletion to full field water injection. The current Ekofisk II redevelopment entails the installation of two new platforms at the Ekofisk complex. One new wellhead platform with 50 slots, and one new Processing platform. During the redevelopment with 50 new wells to be drilled, methods are required to quickly reevaluate well locations based on current reservoir characterization. 18 of the original 50 wells have already been approved for drilling and 11 of these are in production. As the 3D models are updated based on the new results from the completed wells new locations are also be reviewed for drilling. This was the underlying challenge which is addressed in this work. Ekofisk field - optimization code The 3D model is constructed in 33 X 40 X 12 grid and takes approximately 45 min. CPU run time for the forecast case. Because it was the fwst application of the process, simple constraints were introduced during the well generation phases I and 11: 1. Only the grid blocks with a water saturation less than 0.3 could be open to production. 2. At least 4 layersblocks should be completed for each well. Both conditions were chosen on the base of a series of sensitivities runs, with the purpose to avoid an early water breakthrough and to reduce the number of wells and consequently the CPU time required during the 3D simulation runs. No significant changes in the final well locations were demonstrated by starting from a larger number of wells using a water saturation limit lower than 0,3 or reducing below 4 the number of layers open for each wells. Figures 1 shows the simulation grid with the 562 vertical well locations obtained at the fist step of the process. No wells were completed in the mature water flooded areas of the field and in the surrounding aquifer.

SPE 39754

SURVIVAL OF THE FITTEST AN OPTIMISED

WELL LOCATION ALGORITHM

FOR RESERVOIR

SIMULATION

With the purpose to maximize the final field recovery, the were choose equal to the well final ~tness function hydrocarbon recovery expressed in barrels of oil equivalent (BOE). The find number of well locations was set to 32. The remaining 18 wells have previously been approved for drilling. When this number is reached the process automatically ends. The 3D simtiation forecasts were run until December 2030, for each step of the process. a watercut limit of 90% was used as shut-in criteria for the production wells. Figures 2 to 8 show the optimal well locations determined at each of the eight steps of the process. Figure 9 presents the comsponding field recovery factors versus number of wells.

however sensitivities of well locations, injection removingladding injectors were conducted. Because of the high geological uncertainties three different geological models were incorporated in the 3D simulation grid. We refer to them as: Deterministic, classical parameters mapping based on well data; Seismic, result of a seismic Iithology study; Stochastic, one of the most probable realizations of a geostatistical study. Smarbukk field - optimization code Phase I and II, well generation and static screening: 1. Only the grid blocks with a water saturation equal to the irreducible water saturation were open to production. 2. No wells could be located in the two nearest blocks surrounding existing gas injection and production wells 3. At least 4 layers/blocks should be completed for each well, With these constraints, 450 production vertical wells were generated in the 3D simulation grid, avoiding an early water production and gas breakthrough. In addition, a subroutine was written to automatically connect, in the simulation model, each well to the closest subsea template and to assign the corresponding well flow performance table. In this way it has been possible to properly simulate the production constraints associated to each subsea template, like the production start up time, and to run the field production forecasts under surface production limits. Figures 10 shows the Sm@rbukk field simulation grid with the vertical well locations obtained at the fist step of the process. The dynamic well selection phase was run in parallel for each of the six production templates. At the end of the production forecasts the wells belonging to each subsea template were ranked separately, based on a fitness function set equal to the well final oil recovery, during the initial liquid production phase, or equal to a combination of the well final oil and gas recoveries, during the gas export phase. To complete the automatic well location routine 9 steps, inclusive of 3D simulation forecasts, were required fir a total running time of about 20 ~U hours. Using the automatic well location process, 18 new vertical well locations have been defined, over 450 possible locations, for each of the three different geological models, considerable reducing the overall time to perform the simulation study and increasing the net present value of the project, with a high oil and gas recovery acceleration, compared with the previous well locations, manually defined. A plot of hydrocarbon recovery versus number of wells was obtained for the entire field and for every production subsea template allowing an easy display of the incremental recovery associated to each production well and consequently an objective way to define the final number of wells required for the field development. Figures 11 12 and 13 present the final well location generated by the automatic well location process for each of the tree

To verify the location routine several tests were made. Manually moving the automatic optimized well locations in the surrounding grid blocks led, in all cases, to a final recovery lower than that of the optimization code. In addition to this verification an additional case was made starting from the original 50 wells to be drilled in Ekofisk. The 18 wells which had previously been approved and completed all were located within 200-300 meters of the optimum location of the code. Thus not only verifying the methodology used but also providing confidence in past Iocation decisions. This result is actually not entirely surprising as the analysis is based on the same 3D model. The major difference is that using the optimization code the results were available within 24 hours. The manual methodology took approximately 2 man months to obtain the same locations. Using the automatic well location process, 32 vertical wells have been located with a large increase in the 3D model final field recovery compared with the old manual locations. Although, only vertical wells could be generated. It was possible to obtain a good indication about possible horizontal or multilateral well paths identifying the areas with a high production potential from the vertical well location maps. For instance, In figures 7 and 8 several vertical wells, located in nearby grid block cells, could be effectively replaced by a single advanced well. Sm@rbukk field - reservoir description The Sm@rbukk field, discovered in 1985, is located approximately 200 Km of the cost of Mid-Norway. The field production strategy-is divided in two stages. The first stage is a liquid production phase with four subsea production templates. During this phase, most of produced gas is reinfected in the field for pressure maintenance. The second stage is mainly intended for gas export. Two production templates will be added and aIl the injection wells will be converted to producers. A maximum number of four wells can be connected to each subsea template. The reservoir study focused essentially on optimizing the future production well locations for each subsea template and to verify the total number of wells for a complete exploitation of the field. Four drilled wells were considered fixed as the

~=~

G. SANTELLANI,

B. HANSEN AND T. HERRING

WE

39754

different geological models. Due to the large differences in the permeability distribution and absolute values, few production wells were obtained in the same location in all the geological modeIs. The wells with consistent locations are the wells with less associated risks and were recommended to be prioritized in the drilIing list. Discussion The results of this work are of a very practical nature. The well locations which are selected have proven to be correct in the since that the same locations have been selected using traditional methods. Obviomly the locations are only as accurate as the simulation model used to define them, and this wilI remain the weakest link. The objective of improving the efficiency of locating wells has been met leading to increased value of the asset. It is recommended that this procedure be used in connection with geostatistic realizations in order to evaluate well risk and overall reserves in alternative reservoir descriptions. The question of whether or not well locations defined with our automatic routine are truly the optimum in a rigorous sense is valid and has not been addressed. Conclusions 1. An automated technique to optimize vertical production well locations in 3D simulation models as been defined proving to be a powerful and robust tool 2. The time savings of the method is significant and allows well optimization to be utilized on a much larger scale. 3. Verification using traditional methods confirms the validity of the results. 4. Use of the method in various geological realizations leads to a prioritization of low risk ~ells which are consistently obtained in similar locations.

3<

presented at the 1994 SPE International Petroleum Conference Veracruz, Mexico, 10-13 October. Beckner B.L. and Song X.: Field Development Planning Using Simulated Annealing Optimal Economic Well Scheduling and Placement; paper SPE 30650 presented at the 1995 SPE Annual Technical Conference and Exhibition, Dallas, TX, U.S.A., 22-25 October.
and Exhibition,

,,,

. . ....

. ... . . . .

.... .....................m . ....

Acknowledgments The authors wish to thank the management of the Ekofisk and Sm@rbukk fieId partner companies, Phillips Petroleum Co., Fina Expiration Norway SCA., Elf Petroleum Norge AS, Norsk Hydro Production AS, TOTAL Norge, Statiol, Mobil Exploration Norway, Neste Petroleum, Saga Petroleum for permission to publish this paper. The opinions in this paper are those of the authors and do not necessarily represent those of the partner companies. References 1. MrinaI, K. Sen, Akhil Datta-Gupta, Stoffa P,L., Lake L,W. and Pope G.A.: Stochastic Reservoir Modeling Using Simulation Annealing and Genetic Algorithms; paper SPE 24754 presented at the 1992 SPE Annual Technical Conference and Exhibition, Washington, D.C., U.S.A., 47 October. 2. Rian D.T and Hage Asmund: Automatic Optimization of well bcations in a North Sea Fractured Chalk Reservoir Using a Front Tracking Reservoir Simulator, paper SPE 28716

Fig. 1 - Ekofisk field, vertical well location process step 1. 258

well locations.

Automatic

SPE 39754

SURVIVAL OF THE FITTEST AN OPTIMISED

WELL LOCATION ALGORITHM

FOR RESERVOIR

SIMULATION

i,

.,

... . .

,,

.... Fig. 2 - Ekofisk field, vertical well location process step 2. well locations. Automatic Fig. 4- Ekofisk field, vertical well location process step 4. well locations. Automatic

.................i...{~&.........* ..

PWJ

Fig. 3 - Ekofisk field, vertical wetl location process step 3.

well locations.

Automatic

Fig. 5 - Ekofisk field, vertical well location process step 5. 259

well locations.

Automatic

G. SANTELLANI,

B. HANSEN AND T. HERRING

SPE 39754

.+ ,. +

............ ..................

Fig. 6 - Ekofisk field, vertical well location process step 6.

well locations.

Automatic

Fig. 8 - Ekofisk field, final vertical Automatic well location process step 8.

well

locations.

EKOFISK FIELD, AUTOMATIC WELL LOCATION PROCESS

600

500

400

300

200

100

.......................

... , : .:

:..

.. .

WT ~-~.

NUMBER OF WELLS

Fig. 7 - Ekofisk field, vertical well location process step 7.

well locations.

Automatic 260

Fig. 9 - Ekofisk field. Final hydrocarbon recovery, expressed in barrels of oil equivalerit (BOE), and final O-ii recovery versus number of wells.

SPE 39754

SURVIVAL OF THE FITTEST AN OPTIMISED

WELL LOCATION ALGORITHM

FOR RESERVOIR

SIMULATION

,., ... . M
m. .,,
,.

F]g. 10- Smarbukk field, vertical well locations. Automatic w;lI location process step 1. Areas belonging to different subsea production templates.

Fig. 11- Smarbukk field, final vertical well Iocations. As a result of the automatic weil location process applied to the Deferminhtic geological model.

Fig. 12- Smorbukk field, final verticai weli locations. As a result of the automatic well iocation process applied to the Stochmtlc model.

9C 1
,

Fig. 13- Smarbukk field, final verticai well iocations. As a result of the automatic well location process applied to the Seismic model.