SPE 165979

Computer Assisted History Matching - A Field Example

R.H.Lind, HOT Engineering; O.Allottai, Akakus Oil Operations; A.M. Gaaim, NOC Libya; H. Almuallim, Firmsoft
Technologies

Copyright 2013, Society of Petroleum Engineers

This paper was prepared for presentation at the SPE Reservoir Characterisation and Simulation Conference and Exhibition held in Abu Dhabi, UAE, 16–18 September 2013.

This paper was selected for presentation by an SPE program committee following review of information contained in an abstract submitted by the author(s). Contents of the paper have not been
reviewed by the Society of Petroleum Engineers and are subject to correction by the author(s). The material does not necessarily reflect any position of the Society of Petroleum Engineers, its
officers, or members. Electronic reproduction, distribution, or storage of any part of this paper without the written consent of the Society of Petroleum Engineers is prohibited. Permission to
reproduce in print is restricted to an abstract of not more than 300 words; illustrations may not be copied. The abstract must contain conspicuous acknowledgment of SPE copyright.

Abstract

Computer assisted history matching has received much interest in the recent past. Several commercial programs are now
available. We have tested some of these programs – most did not meet our expectations and we discontinued using them.
The latest software tested helps in fine-tuning individual well behavior in a simulation model that has been globally
matched. The program identifies model cells which have the highest influence on individual well behavior and modifies
permeability in three directions (and porosity) such that the match is improved. Maximum allowed changes of permeability
(and porosity) are controlled by the user. The procedure is repeated 10 to 20+ times until the match is considered satisfactory
or the model no longer improves.
A flow simulation model is globally matched when all wells can flow historical rates, the pressure history in different
compartments or layers can be reproduced, external aquifer properties are adjusted, and fluid and rock properties are verified.
In such a model computed well watercut can be grossly in error, some wells are too high others too low. Manual fine-tuning
of individual well behavior is a lengthy and frustrating process. We have tested the new computer assisted history matching
program on several models. Improvements to most well matches were spectacular.
Model set up and history match procedures used are outlined on a moderately complex reservoir containing more than 1
billion STB oil initially in place. Production history spans 14 years, current recovery is ~25% of OOIP. Some 40 wells are
active, four of them peripheral injectors. Injection voidage replacement is ~30%; there is a strong bottom and edge aquifer.
The reservoir consists of 3 geological layers; the upper one has a permeability of several Darcy; the middle layer is highly
heterogeneous and forms a partial seal to the lower layer with permeability of ~100 mD. We show examples of individual
well matches before and after computer assisted fine tuning. We also discuss limitations of the program used and manual
interferences required for selected wells.

Introduction
History matching is the process of adjusting a geological model such that it can reproduce measured production and pressure
data. Historical well liquid rates are input in the model for oil reservoirs. We do not normally recommend inputting well oil
rates. Parameters like oil, water, gas rates, watercut, GOR, flowing and static pressures are computed. In gas or condensate
reservoirs historical well gas rates are input in the model.
No matter how much effort went into geological model building the computed production or pressure parameters never
agree with reported ones. During conventional manual history matching selected parameters are modified using a trial and
error approach in an attempt to reproduce reported data. Parameters modified should be those that are most uncertain,
typically external aquifer properties (size and strength) and permeability distribution in the gridded model. Other uncertain
parameters are fault transmissibility, relative permeability curves, dual porosity/permeability parameters, effects of grid size
(resolution) etc. Having this many parameters to tune, history matching is subjective and the results are not unique. State of
the art geological software easily allows building numerous equally likely models, none of which will reproduce reported
history and ideally several geological models should be matched.
What are frequently overlooked are the errors in measured and allocated history data. Water production need not originate
from the reservoir but can come from mechanical problems (casing leaks, poor cement bond).
During recent years several programs have become available to aid in history matching. Initial claims of automatic history
matching are getting quiet, but computer assisted history matching is getting popular. Numerous papers are being published
2 SPE 165979

and various numerical methods are proposed. Rwechungura et al. (2011) summarize several methods used for computer
assisted matching. In this paper we discuss our method of generating flow simulation models using a real field example. We
use a commercially available program for computer assisted history matching FIRMSOFTTECH (2010). The procedure
explained has been used successfully for numerous reservoir studies.

B-Field Field Summary

Initial oil in place exceeds 1 billion STB. Artificial lift is required (ESPs). Edge and bottom water influx is strong, but
peripheral water injection is needed to maintain the desired pressure. Voidage replacement ratio is ~30%. The reservoir is
highly undersaturated and the bubble point pressure will not be reached. Gas saturation is therefore not included in the model,
reducing computing time. Despite the low GOR the oil viscosity is low and the mobility ratio is favourable.
The reservoir has been operating for more than 14 years and the decline phase has started. Current recovery is ~25%. A
few wells are watered out and many started producing significant amounts of water. The objective of this study was to extend
the oil rate plateau and identify optimum infill drilling locations for both producers and injectors. An integrated reservoir
study was made. All field data have been reinterpreted: 3D seismic, core descriptions and measurements, open and cased hole
logs, transient tests, material balance etc. Overall data quality is good. Use of some of the geological data is summarized by
Cubitt et al. (2011) and Gruber et al. (2011).
The reservoir consists of three distinct geological layers; the uppermost unit has a permeability of several thousand mD,
but despite the high permeability its behaviour is single porosity/permeability. The middle layer is highly heterogeneous and
has a continuous shale layer at its base. This middle layer forms a partial barrier to the bottom unit with a permeability of
around 100 mD. Production performance of the few wells which only encountered the middle unit is poor. Figure 1 is a
representative cross-section of the flow model. The upper two units pinch out towards the south. Bottom aquifer layers are
upscaled as vertical resolution is not required for flow simulation.

Upper  Unit 200x200  m  cells

34  active  layers
100x100  m  cells
67  active  layers
18  foot  thickness 9  foot  thickness
– Extremely  high  permeability   96  000  active  cells 442  000  active  cells
(several  Darcy) 10  minutes  CPU  time 1-­‐2  hours  CPU  time

Middle  Unit
– Poor  reservoir,  tight
– Very  heterogeneous
Vertical  exaggeration:  20x
– Continuous  shale  at  base
– Partial  pressure  barrier

Lower  Unit
– Moderate  permeability
– Below  OWC  in  north

Figure 1: Representative Cross Section Through Flow Figure 2: Coarse (left) and Fine Grid Models, Structure
Simulation Model Top

Model Set-up and History Matching

The model grid was defined jointly by geolocial modeler and reservoir engineer in order to have suitable grids for both
geological and flow modeling. For flow simulation we prefer to start with a very coarse grid, sized such that it can be run
through the history period in a few minutes and 3D properties and crossections can be quickly displayed. The purpose of the
coarse model is to understand global field behaviour, identify pressure compartments and impact of faults or barriers, adjust
global permeability and external aquifer properties. We match the pressure history of the coarse model, but do not attempt
watercut matching nor make predictions. Simulation experts frequently recommended setting up a material balance model
before starting flow simulation. Except for the simplest reservoirs containing a few wells in a single pressure compartment,
we are not in favour of material balance modeling. These models are too simplistic, at least for the reservoirs we normally
study. We prefer coarse flow models instead. The coarse grid should be set up early during geological modeling. If faults can
be modeled with a coarse grid, they can also be modeled with a finer grid when typically one coarse cell is areally represented
by 4 fine cells. Figure 2 compares coarse and fine grids used.
During integrated reservoir studies we start flow simulation well before geological modelling is finalized. The purpose is
to identify problems early. In the present study two such problems were identified and could be solved easily. The two upper
geological units pinch out in parts of the reservoir. In the first model the middle unit disappeared before the upper one. This
resulted in direct communication between high permeability upper and lower units and resulted in identical pressure; clearly
contradicting pressure history. It was easy to fix the grid early rather than using transmissibility multipliers in the flow model
later on. The reservoir is bound by a fault zone along one edge. It is not a single fault, but a series of faults with small throws.
SPE 165979 3

This is the direction were the upper layer receives the edge water drive. In the initial model some faults had a larger throw
than the reservoir thickness, thereby eliminating flux in the uppermost layer. While still honouring seismic interpretation, fault
throws were reduced slightly and the tilt of intermediate blocks increased. This allowed lateral water influx in the high
permeability upper unit without defining non neighbor connections or a fracture zone to model flow up the fault plane, which
most simulators cannot do easily.

Horizontal Permeability vs. Porosity Maximum  Allowed  Permeability  Changes

10000
– Maximum:  10x
Horizontal Core Permeability [mD]

1000 – Minimum:    0.1x

– PermX/PermY <  20
100
– PermY/PermX <  20  
10 Maximum  Allowed  Porosity  Changes
1
– Maximum:  1.01x
Upper Unit
– Minimum:  0.99x
0.1 Middle Unit

Lower Unit Table 1: Permeability and Porosity Constraints for

0.01
0.00 0.05 0.10 0.15 0.20 0.25 Computer Assisted Matching
Core Porosity [ ]

Figure 3: Core Plug Permeability versus Porosity

Main parameters modified during coarse grid matching were analytical aquifer properties and permeability at the base of
the middle layer. Different (Carter-Tracy) aquifers were attached to high permeability upper and moderate permeability lower
units. Some infill wells drilled after production start encountered watered out zones. This was also reproduced in the coarse
model. It turned out that aquifer influx is essentially from one direction only. Numerical tracers were used to track the
injection water and external aquifer water movement. Formation tester (RFT) measurements show a distinct pressure change
across the middle unit. These formation tester pressures were also matched. Flow meter log results were matched by
adjusting layer skin factors or local permeability. Changes made to the coarse model were introduced in the fine model and
some small manual adjustments made. This model does reproduce the pressure history in the various areas of the field and the
overall water influx behavior is correct, but individual well watercuts can be grossly in error. Manual well watercut matching
tends to be very time consuming and frustating.

Computer Assisted Matching

In the present example, computer assisted matching was used to fine tune the computed well watercuts and at the same
time preserve the static pressure match. In this reservoir it is tempting to increase permeability of the tight streaks to increase
watercut of selected wells, but this would upset the pressure match. Water influx to the wells in the upper layers does not
come directly from below, but from the edge. Static well pressures have to be taken into account by the history matching
software.
In the assisted history matching program employed here, the mismatch at a given well is quantified by summing up the
squared difference between measured and computed values. The overall mismatch is then computed as a weighted sum of
individual well mismatches, where the weights are determined by the user. The overall mismatch quantified this way is taken
as the “objective function” which the program attempts to minimize. The program uses a variant of the “adjoint method” to
compute the derivative of this objective function with respect to every parameter at each cell. Changes in grid properties that
improve the match are then computed based on these derivatives. More details can be found in Almuallim et al (2010) and
FIRMSOFTTECH (2010).
Parameters modified by the program are permeability (independent in 3 directions) and porosity. Box modifiers are not
used. The program identifies individual cells which impact well performance and makes changes to these cells in an iterative
process. Constraints on the maximum changes allowed are defined by the user and are summarized in Table 1. Original
permeabilities in the model are assigned to individual cells based on data from the core porosity-permeability plots (Figure 3).
Frequently the permeability varies by 2 to 3 orders of magnitude for the same porosity. It is therefore sensible to allow
changes to permeability in this same order of magnitude. We did set the maximum allowed model permeability to 10 000 mD.
Outputs of the program are ‘include’ files of permeability (in three directions) and porosity, which are used for another
simulation run.
Match quality tends to increase rapidly with the first iterations, but then slows down. When matching watercut it is
necessary that the well produces some water, otherwise the program cannot identify possible changes as no derivative can be
computed. The opposite is not true. If the model well produces water, but the real well doesn’t, the program has repeatedly
reduced the computed watercut to zero.
Figure 4 compares permeability from the geomodel on the left with matched model permeability in a selected layer in the
oil leg. The changes introduced by the assisted history matching program are truly minimal; applied only as and where needed
4 SPE 165979

to improve the match. These differences are hence hard to see. It is obvious that the changes did not alter the geological
concept, yet the production/pressure history is matched.
One limitation of the program is its inability to improve water producing wells where the model does not compute any
water at all. This was happening at one of our wells, where we increased permeability locally using a box multiplier, see
arrow. This did draw some water to the well and the program achieved a perfect match afterwards.

Upscaled Matched Matched

Horizontal Perm. PermX PermY

Figure 4: Horizontal Permeability before and after Computer Assisted Matching

Figure 5 shows permeability histograms of the geomodel and the matched model for the upper and lower units. Changes
made during computer assisted history matching are small and well within the uncertainty of the data. Figures 6 compares
measured and computed data for the whole field after global matching. On a field basis this would be acceptable, but
individual well behavior can be grossly in error, but cancel each other out. Figure 7 is the field plot after computer assisted
matching.
After computer assisted matching some more manual tuning was done in selected wells to adjust the flowing bottom hole
pressure. Flowing pressure was not included into assisted matching, only static pressure. Flowing pressure adjustments were
made using a productivity index multiplier. Modifying the skin factor would have given the same result, but changing the PI
multiplier is simpler for the software used.
Figures 8 to 25 compare well plots before and after computer assisted matching. A series of plots were chosen that show
the variation of well watercuts that have been matched, without destroying the pressure match.

Conclusions
Using a field example we show our approach for flow modeling. We start with a coarse grid model which we match
manually for global behavior, but not individual well behavior. What has been learned in the coarse model is introduced in the
fine grid model. Computer assisted matching is then used to fine tune individual well matches. Results of the computer
assisted well matches have been spectacular.
We have used this program in several studies. Significant improvements were usually achieved, and in situations where
this is not the case, this would be an alert indicating some global problems such as:

The flow model is too coarse.

The geological model needs to be reviewed.

Reported or allocated production data are erroneous.
In the particular model presented here, the geological model is good and the reported data are largely error free, and hence the
spectacular improvement of most individual well matches.

Acknowledgement
The authors wish to thank Akakus Oil Operations for the permission to publish data.
SPE 165979 5

Upper  Unit
Upscaled Matched Matched
Horizontal Perm. PermX PermY

Lower  Unit
Upscaled Matched Matched
Horizontal Perm. PermX PermY

Figure 5: Horizontal Permeability Histograms before and after Computer Assisted Matching

References

Almuallim, H., Edwards, K.and Ganzer L. (2010): History-Matching With Sensitivity-Based Parameter Modifications at
Grid-Block Level, SPE-131627.

Cubitt, C.J., Gruber, W., Stummer, B.C. and Allottai, O. [2011]: Depositional Environments of the
Mamuniyat Formation in the Central Murzuq Basin, Libya., 73rd EAGE Conference.

Gruber, W., Cubitt, C.J., Stummer, B.C. and Allottai, O. [2011]: Reservoir Characterization of the Mamuniyat Formation
(Murzuq Basin, Libya) - An Integrated Approach from Core to Facies, 73rd EAGE Conference.

Rwechungura, R., Dadashpour, M. and Kleppe, J. (2011): Advanced History Matching Techniques Reviewed, SPE-142497.

FIRMSOFTTECH (2010): SenEx User Manual; www.firmsofttech.com.

6 SPE 165979

Liquid rate (input)

Oil rate
Water injection
Watercut

Water rate

Static Pressures

Figure 6: Computed Full Field History, Manual Matching

Figure 7: Computed Full Field History, Computer Assisted Matching

SPE 165979 7

Figure 8: B03 Computed Well History, Manual Matching

Figure 9: B03 Computed Well History, Computer Assisted Matching

8 SPE 165979

Figure 10: B04 Computed Well History, Manual Matching

Figure 11: B04 Computed Well History, Computer Assisted Matching

SPE 165979 9

Figure 12: B08 Computed Well History, Manual Matching

Figure 13: B08 Computed Well History, Computer Assisted Matching

10 SPE 165979

Figure 14: B15 Computed Well History, Manual Matching

Figure 15: B15 Computed Well History, Computer Assisted Matching

SPE 165979 11

Figure 16: B19 Computed Well History, Manual Matching

Figure 17: B19 Computed Well History, Computer Assisted Matching

12 SPE 165979

Figure 18: B21 Computed Well History, Manual Matching

Figure 19: B21 Computed Well History, Computer Assisted Matching

SPE 165979 13

Figure 20: B22 Computed Well History, Manual Matching

Figure 21: B22 Computed Well History, Computer Assisted Matching

14 SPE 165979

Figure 22: B23 Computed Well History, Manual Matching

Figure 23: B23 Computed Well History, Computer Assisted Matching

SPE 165979 15

Figure 24: B30 Computed Well History, Manual Matching

Figure 25: B30 Computed Well History, Computer Assisted Matching