You are on page 1of 2

A Data-Driven Model

for History Matching and Prediction

I n this paper, the authors derive and


implement an interwell numerical
sufficiently well that, when INSIM is
used for history matching, the result-
2

simulation model (INSIM) that ing history-matched model can be en- (Ti,2 ,Vp,i,2) (Ti,1 ,Vp,i,1)
can be used as a calculation tool to tered into INSIM to provide reasonable
approximate the performance of a future predictions and provide informa-
reservoir under waterflooding. In tion on the flow dynamics of the res-
INSIM, the reservoir is characterized ervoir. It is hoped that the model and
1
as a coarse model consisting of a methodology presented here will prove i
number of interwell control units, useful for monitoring and understand-
and each unit has two specific
parameters: transmissibility and
ing waterflooding operations conducted
on a black-oil reservoir and that INSIM
(Ti,3 ,Vp,i,3)
3
control pore volume. INSIM is will ultimately be useful for waterflood-
applied to perform history matching ing optimization. Fig. 1—Illustration of modeling
for parameter estimation and to Models based on the statistical cor- of volume-flow-unit connections
infer interwell connectivity and relation or the connectivity between in- between wells. Ti=transmissibility
of Node i.
geological characteristics. jectors and producers estimated from
flow-rate data have been used previ-
Introduction ously to characterize reservoirs for the to injectors. At the same time, INSIM
History matching with a reservoir simu- purpose of waterflooding management. retains the computational efficien-
lator is the most common way to con- Unlike previous correlation-based mod- cy of previous correlation-based mod-
dition rock-property fields to produc- els, INSIM is able to effectively predict els that incur far less computational
tion data. However, production data are the water cut and oil-production rate cost than a traditional numerical reser-
never sufficient to resolve the reservoir and hence can be used as the forward voir simulator.
properties (e.g., gridblock permeabili- model for assisted history matching of
ties), and few assisted-history-matching these data. Specifically, the model can INSIM
tools exist in commercial reservoir sim- be used in automatic history match- In INSIM, the reservoir is first con-
ulators. Consequently, when a reservoir ing. Moreover, the model is derived di- sidered as a network model consist-
simulator is used as the forward model rectly from the correct two-phase-flow ing of a series of units connecting well
when history matching, the number of mass-balance equations, and thus the pairs. Unlike correlation-based models,
reservoir parameters is often reduced transmissibilities derived from histo- INSIM allows not only for injector/pro-
to a small number on the basis of com- ry matching reflect an average trans- ducer connections but also for injector/
putational experiments and physical missibility between wells. In addition, injector and producer/producer connec-
insight. Although the INSIM method- because INSIM is based on simulation tions. Thus, it is expected that INSIM
ology introduced here does limit the flow equations, it can incorporate large can better resolve the flow that occurs
number of history-matching parame- changes in flow rates, flow directions, in the region between pairs of wells of
ters, the primary objective of INSIM is and injector allocation factors—the the same type.
to provide a fast, simplified simulation interaction between pairs of produc- As shown in Fig. 1, Well Node i is as-
model to calculate flow and transport ers and the conversion of producers signed a volume denoted by Vp,i depict-
ed by the dashed red circle. At present,
This article, written by JPT Technology Editor Chris Carpenter, contains highlights only fully penetrating vertical wells and
2D flow are considered; thus, in Fig. 1,
of paper SPE 173213, “INSIM: A Data-Driven Model for History Matching and
Vp,i is the cylindrical volume enclosed
Prediction for Waterflooding Monitoring and Management With a Field Application,”
by the red-dashed circle extended over
by Hui Zhao, SPE, Yangtze University; Zhijiang Kang, China Petroleum and Chemical the reservoir thickness. Flow in the in-
Corporation; Xiansong Zhang, China National Offshore Oil Corporation; Haitao Sun terwell area between Well i and its con-
and Lin Cao, Yangtze University; and Albert C. Reynolds, The University of Tulsa, nected well nodes is modeled as flow
prepared for the 2015 SPE Reservoir Simulation Symposium, The Woodlands, Texas, within a region characterized by two
USA, 23–25 February. The paper has not been peer reviewed. parameters, transmissibility and a con-

For a limited time, the complete paper is free to SPE members at www.spe.org/jpt.

JPT • APRIL 2016 81


trol pore volume, where the transmissi- ration tracking for each control unit can ters represent the reservoir parameters
bility controls flow between Well j and be viewed as a 1D problem. For oil/water (optimization parameters or history-
Well i in the interwell volume. For Fig. 1, two-phase flow, Buckley-Leverett theo- matching parameters). In general, the
j=1,2,3, but there is no limit to how ry provides a relationship between the inverse problem is ill-posed and some
many wells can be connected to Well i. cumulative water injected and the loca- regularization is required. However, be-
However, the connections have to be de- tion of the flood front. In the tradition- cause of the limited number of con-
fined before any computations can be al way, Buckley-Leverett theory is used nections used to date, the authors have
performed. For each single Well Node for saturation tracking from injector to been able to generate reasonable results
i, there is a well-controlled pore vol- producer. In the computation of the sat- without regularization.
ume such that the summation of all Vp,i uration distribution in INSIM, standard Because the computational time re-
should be equal to the pore volume of Buckley-Leverett theory is applied if the quired to run INSIM is small, the au-
the reservoir. Practically, one can gen- flow direction along a connection be- thors simply use a gradient-based op-
erally use the area enclosed by link- tween a well pair does not change. How- timization algorithm with derivatives
ing all the middle points between Well i ever, for the case where flow reversal computed by the finite-difference meth-
and its connected adjacent wells to de- occurs—for example, because of con- od to optimize the objective function.
fine Well i’s controlled area and hence version of a producer to an injector— The algorithm used is an interior-point
define Vp,i. the Buckley-Leverett computations are method from commercially available
The connections between wells must applied in an ad hoc manner to track the optimization tools. In the complete
be specified a priori in INSIM. The com- flow of injected water. Buckley-Leverett paper, the authors investigate the vi-
plete paper contains a discussion of the theory is also used to track flow from ability of INSIM by considering two
material-balance equation and the pres- producer to producer. This last option is examples, a synthetic example and a
sure calculation. useful when two producers (P-a and P-b) field example.
are connected, but only one of these
Saturation Tracking and Well- producers—say, P-a—is connected to Conclusions
Dynamic-Data Prediction. Because the a certain water injector, I-a. In this sce- Despite the ad hoc computation of satu-
reservoir is modeled as a series of con- nario, injected water flows along the ration at well nodes after flow-direction
trol units between well pairs, the satu- connection from I-a to P-a until break- reversals, the computational results
through. After breakthrough, if all of suggest that INSIM is a simple, compu-
the water cannot be produced by P-a, tationally efficient forward model that
in the INSIM model, some of this water can replace a reservoir simulator when
will flow along the connection from P-a history matching phase-rate data or
to P-b if the pressure at P-b is lower than water-cut data obtained during water-
Changing Your the pressure of P-a. flooding operations. Although the re-
sults illustrate the validity of the INSIM
Address? Well-Rate-Allocation Factors. Results model, it may be possible to improve
from INSIM can be used to calculate this model significantly by adding a
Let SPE know.
the allocation factor for the fluid rate more rigorous front-tracking computa-
+1.972.952.9393
between each injector/producer well tion and by the use of the parameters of
pair by use of the same basic proce- power-law relative permeability curves
dure used in correlation-based mod- as history-matching parameters.
els. However, unlike the results from The INSIM process for solving for
Update Your correlation-based methods, the alloca- pressure and saturation is fast and sta-
tion factor changes with time and thus ble. History-matched results obtained
Member Profile reflects the dynamic interactions be- by the combination of INSIM with an
tween wells. For an injector/producer optimization algorithm allow one to de-
http://www.spe.org/ pair, the allocation factor of the liquid- termine the formation characteristics
members/update flow rate from Injector i is defined as between well pairs, the rate allocation
the liquid flow from Injector i to the pro- factors for each injector (the fraction
ducer divided by the total injection rate of injected water that flows to each pro-
of the injector. ducer connected to the injector), and
the principal directions of the flow of
SPE Benefits History Matching Phase-Rate water from each injection well. Unlike
or Water-Cut Data correlation-based models, INSIM can
Discover the possibilities.
Predictions with INSIM depend on produce reasonably accurate phase-flow
http://www.spe.org/ the sets of the two parameters defined rates and water cut and handle chang-
members/benefits for each connected well pair. In histo- es in flow directions caused by closing
ry matching the phase-rate or water- wells or by the conversion of producers
cut data at producers, these parame- to injectors. JPT

82 JPT • APRIL 2016

You might also like