You are on page 1of 9

SPE-197585-MS

Water Flooding Performance Prediction in Layered Reservoir Using Big Data


and Artificial Intelligence Algorithms

Cunliang Chen, Ming Yang, and Xiaodong Han, Tianjin Branch of CNOOC, China Co., Ltd; Jianbo Zhang, China
University of Petroleum, East China

Copyright 2019, Society of Petroleum Engineers

This paper was prepared for presentation at the Abu Dhabi International Petroleum Exhibition & Conference held in Abu Dhabi, UAE, 11-14 November 2019.

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
Managing oil production from reservoirs to maximize the future economic return of the asset is an important
issue in petroleum engineering. One of the most important problems is the prediction of water flooding
performance. Traditional strategies have been widely used with a long run time and too much information
to solve this problem. Therefore, it is urgent to form a fast intelligent prediction method, especially with the
development of large data processing and artificial intelligence methods.
This paper proposed a new method to predict water flooding performance using big data and artificial
intelligence algorithms. The method regards layered reservoir as a vertical superposition of a series of
single layer reservoirs. An injection-production analysis model is established in each single layer reservoir
respectively. And then a superposition model is established only by production data and logging tools data.
Finally, the least square principle and the particle swarm optimization algorithm are used to optimize the
model and predict water flooding performance.
This method has been tested for different synthetic reservoir case studies. The results are in good
agreement in comparison with the numerical simulation results. The average relative error is 4.59%, but
the calculation time is only 1/10 of that of numerical simulation by using artificial intelligence method. It
showed that this technique has capability to predict water flooding performance. These examples showed
that the use of artificial intelligence method not only greatly shortens the working time, but also has a higher
accuracy.
By this paper, it is possible to predict the water flooding performance easily and accurately in reservoirs.
It has an important role in the field development, increasing or decreasing investment, drilling new wells
and future injection schedule.

Introduciton
Oilfield development index prediction is very important to oilfield development plan design and adjustment
(Anh P. Nguyen, 2011; Martín Salazar, 2012; A.AI Saidi, 2015; Fei Cao, 2015; Hui Zhao, 2015). It is also an
important research subject of reservoir engineering. Analyzing the law of dynamic performance of reservoir
development and predicting the future trend of reservoir performance are useful for reservoir managers
2 SPE-197585-MS

to master the law of oilfield development in time and adjust the development plan, and improve effect of
oilfield development. It is of great practical significance to enhance the ultimate recovery factor of oilfield.
In order to accurately predict the change process and evolution trend of oilfield dynamics, a lot of research
work has been done and many prediction methods have been put forward by petroleum workers. Generally,
it can be divided into four categories: empirical formula method, mechanism model method, numerical
simulation method and artificial intelligence method. Empirical formula method includes decline curve
method, Weng's curve method, logistic prediction method, water flooding characteristic curve method,
statistical formula method and the combination of these methods. Decline curve method and water flooding
characteristic curve method are the most widely used methods. J. J. Arps (1945) firstly proposed the
decline curve method. According to the different percolation index, the decline curve can be divided into
three types: exponential, harmonic and hyperbolic formulas. Under the stable working system, the future
oil production can be predicted by reasonable regression of historical production. There are nearly 100
kinds of water drive curve methods. By properly combining the parameters of oil field such as liquid
production, oil production, water production or water-oil ratio, certain mathematical formulas can be
established by fitting them. The future development indexes can be predicted on the basis of precise data
fitting (Yuanqian Chen, 2011). In recent years, many researchers found that the fitting accuracy of water
flooding characteristic curve is low at the stage of high water cut (Jinqing Zhang, 2019). They improved
the fitting accuracy by constructing different curves and achieved certain results. The mechanism model
method is a method based on the mathematical model. The mathematical model is established with reservoir
engineering and seepage mechanics theory, solved by analytic method, and applied to the prediction
of development indicators. The numerical simulation method predicts future development indicators by
establishing accurate geological models and simulating. The above three methods are commonly used in
the early research, which have a certain theoretical basis. But the methods have disadvantages of complex
application, multiple restrictions, and low prediction accuracy. However, with the continuous improvement
of artificial intelligence method and large data method, the application of artificial intelligence method
in oilfield development is more popular. Xu Shaohua (2015) used artificial neural network method to
forecast development index. Zhao Ling (2018) established prediction model of oilfield development index
based on process support vector regression method. Zhang Dongmei(2019) used the EEMD Gauss process
autoregressive model to predict the development indicators. These methods are relatively simple and have
a certain accuracy. But the theoretical basis is insufficient. On the basis of previous studies, theoretical
research and artificial intelligence method are both used in this paper. A new method for developing dynamic
index prediction is proposed. Firstly, the connectivity model between wells in multi-layer reservoirs is
established. Then the development index is predicted combing with Gentil's empirical formula.

Interwell connectivity model for multilayer reservoirs


Multilayer reservoirs are essentially multiple oil-water systems. When ignoring interlayer cross-flow, a
multilayer reservoir can be regarded as the vertical superposition of a series of single-layer reservoirs.
According to the material balance equation of elastic-water drive reservoir, the equation can be expressed
to the layer j as:
(1)
Eq. 1 can also be expressed as:
(2)

Where: .
At the time of (t-1), we obtain:
SPE-197585-MS 3

(3)
At the time of (t), we obtain:
(4)
Combine Eq. 4 and Eq. 5, we can have
(5)
Similarly, for the production unit where the production well is located, there are
(6)
When the production well is affected by multiple injection wells in the layer j, the water injection in the
layer j is expressed as

(7)

The liquid production in the layer j can be expressed as

(8)

If water injection wells have water injection profile or separate layer injection, and production wells
have fluid production profile or separate fluid producing data, Eq.8 can be transformed into a relatively
simple multiple linear regression model to solve. In reality, the data of water injection profile and separate
layer injection data are easy to obtain. But due to the limitation of geological and economic conditions,
production profile or separate fluid producing data are deficient. So the calculation conditions for Eq. 8
are infeasible. Because the production data of the whole well is easy to obtain, the calculation model is
improved by introducing the well production data.
The fluid production of well n is the superposition of the fluid production of different perforated layers
(as shown in Figure 1).

Figure 1—Fluid production mechanism in multi-layer reservoirs


4 SPE-197585-MS

(9)

Prediction of production index


The wells connectivity model shows that the fluid production of a well is the superposition of the fluid
production of different perforated layers. The fluid production of each layer is the affected by natural energy
and artificial injected water flooding energy. By ignoring natural energy, the liquid production comes from
the contribution of water injection. According to the water injection plan, the future liquid production can
be predicted.
Gentil (2005) proposed an equation to express the relationship between the water-oil ratio and cumulative
water injection according to the law of energy conservation.
(10)

(11)

When injection and production are in balance, cumulative water injection is equal to the sum of
cumulative oil production and cumulative water production. The oil content can be expressed as:

(12)

Oil production is the total liquid production multiplied by the oil content.
(13)
Oil production can be calculated for any producer by combining Eq. 11 and Eq. 13 as:

(14)

For multi-layer reservoirs, the production index of each layer can be obtained by calculating the oil
production with Eq. 14 for each layer.

Model solution
Injection-production connectivity coefficient is the proportion coefficient of the water injection volume that
acting on fluid production in production wells to the total water injection volume So the value of injection-
production connectivity coefficient in each layer is between 0 and 1. And the sum of the corresponding
well production affected by the injection well is not greater than the total injection volume. The sum of
all injection-production connectivity coefficients associated with the injection well is not greater than 1,
as shown in Eq. 10.

(15)

Eq. 9 and Eq. 15 is difficult to solve directly, so the optimization algorithm based on the least squares
principle is proposed to solve the problem. According to the principle of least squares, the model in this
paper is transformed into an optimization model with constraints.
SPE-197585-MS 5

(16)

The particle swarm optimization (PSO) used to solve the above model. PSO starts from stochastic solution
and searches for the optimal solution by iteration. Although it also evaluates the quality of solution by
fitness, it is simpler than genetic algorithm rules. In recent years, PSO has shown its superiority in solving
practical problems with its advantages of easy implementation, high accuracy and fast convergence. The
process of solving the model by PSO method is shown in Figure 2.

Figure 2—Flow chart of particle swarm optimization method

Field application
SZ36 offshore oilfield is located in the lower Liaohe Depression of Liaodong Bay and the middle part of low
swell in Liaoxi. It is lacustrine delta deposit. The reservoir type is stratified structural reservoir. Directional
wells are used to develop the oilfield. At present, it has been developed for more than 20 years and is at the
moderate and high water cut development stage. The study of dynamic connectivity in multi-layer reservoirs
is of great significance to tap the potential of remaining oil in the later stage.
The feasibility of the method is applied and verified in X well area. The X well group in this well area
is a five-point well pattern. The injection well X9 injects water to layer I and layer II. The water cut of the
production well is high. In order to monitor the water injection effect of each layer in the production well
group, the stratified tracer test was implementation. The trace element α was injected into Layer I and the
trace element gamma γ was injected into Layer II. The monitoring results showed that the injected water in
layer I dash to X4 and X8 more fast, but in layer II, the injected water dash to X1 and X4 more fast.
The injection-production connectivity of layer I and layer II is also calculated. As shown in Figure 3, the
arrow in the figure points from the injection well to the production well. The length of arrow represents the
value of injection-production connectivity coefficient. The injection-production connectivity between wells
is described intuitively. The calculation results are in good agreement with the tracer monitoring results,
which verifies the feasibility of the method in this paper. According to the calculation results, profile control
6 SPE-197585-MS

and flooding operations were designed and implemented to the two layers respectively. After the operation,
the water cut of the well group is reduced by 2%, and the daily oil production increased 23m3.

Figure 3—Schematic diagram of calculation results for injection-production connectivity of well group X9

Table 1—Propulsion velocity of oil-water front from well X9

Tracer breakthrough Propulsion velocity of


Injection Production Well time(d) oil-water front(m/d)
well well space(m)
I (α) II (γ) I (α) II (γ)

X1 310 10 8 31.0 38.8


X4 335 8 8 41.9 41.9
X9
X5 320 10 12 21.0 26.7
X8 330 8 10 41.3 33.0

The prediction of oil well dynamic index is carried out. As shown in Figure 4~7, each well has entered a
rapid decline stage of production. The strategy of increasing liquid production in the later stage is proposed
to stabilize the production of oil wells.

Figure 4—Actual production data and prediction data of well X1


SPE-197585-MS 7

Figure 5—Actual production data and prediction data of well X5

Figure 6—Actual production data and prediction data of well X8

Figure 7—Actual production data and prediction data of well X9


8 SPE-197585-MS

Conclusion
1. The multi-layer reservoir is regarded as the vertical superposition of single-layer reservoirs. Based
on the principle of material balance, a new quantitative evaluation method of dynamic connectivity
between wells in multi-layer reservoirs is established by using injection and production data. A
prediction method of development index is proposed.
2. The least squares principle and artificial intelligence optimization algorithm are used to solve the
computational model automatically. The difficulty of solving overdetermined equations directly is
solved.
3. The results of field application calculation are consistent with those of tracer stratification test. The
feasibility of the method is verified. This method offers a guiding significance for oilfield stratification
treatment. By predicting the future development index of oil wells, the future of oil wells can be
foreseen, and the development strategy can be adjusted in time to improve the ultimate recovery factor
of oil field.

Nomenclature
the accumulated oil production of the layer j;
the crude oil volume coefficient of layer j;
the cumulative water production of the layer j;
the comprehensive compression coefficient of the layer j;
the initial crude oil volume coefficient of layer j;
Nj the geological oil reserves of the layer j;
Δpj the pressure drop of the layer j;
the cumulative water injection volume of the layer j;
the water influx of the layer j;
the reservoir liquid production of the layer j;
the fluid production of non-water injection contribution in the layer j;
the reservoir water injection;
the reservoir instantaneous liquid production of the layer j;
the reservoir instantaneous liquid production of non-water injection contribution in
the layer j;
I the reservoir instantaneous water injection of the layer j;
j

n the order number of production well;


the reservoir instantaneous fluid production of the layer j for oil production well;
the reservoir instantaneous fluid production for oil production well of non-water
injection contribution in the layer j;
Ij,n the reservoir instantaneous water injection of production unit in the layer j;
m the order number of injection well;
Nj,n the number of injection wells acting on production well n in the layer j;
the connection coefficient between injection well m and production well n in the
layer j;
the instantaneous water injection of injection well m in the layer j;
the liquid production of well n;
SPE-197585-MS 9

NL,n the producing layer number of well n;


R the water-oil ratio;
WI the cumulative water injection;
a,b the fitting parameters;
Nk the total number of water injection wells;
Nt the selected time steps;
Nj,m the number of production wells affected by injection well m in the layer j;
the actual instantaneous reservoir fluid production for production well n.

References
A. AI Saidi, P. Pourafshary, M. AI Wadhahi. 2015. Application of Fast Reservoir Simulation Methods to Optimize
Production by Reallocation of Water Injection Rates in an Omani Field. Paper 172633 presentated at the SPE Middle
East Oil & Gas Show and Conference held in Manama, Bahrain, 8-11 March.
Anh P. Nguyen, Leon S. Lasdon, Larry W. Lake, et al. 2011. Capacitance Resistive Model Application to Optimize
Waterflood in a West Texas Field. Paper 146984 presentated at the SPE Annual Technical Conference and Exhibition
held in Denver, Colorado, U.S.A., 30 October – 2 November.
Chen Yuanqian, Zou Cunyou, Zhang Feng. Application of water drive curve method in oilfield development evaluation
[J]. Fault-Block Oil and Gas Field, 2011, 18 (6):769–771.
Fei Cao, Haishan Luo, Larry W. Lake. 2015. Oil Rate Forecast by Inferring Fractional Flow Models from Field Data.
Paper 173315 presentated at the SPE Reservoir Simulation Symposium held in Houston, Texas, USA, 23–25 February.
Hui Zhao, Zhijiang Kang, Xiansong Zhang, et al. 2015. A Data-Driven Model for History Matching and Prediction for
Waterflooding Monitoring and Management with a Field Application. Paper 173213 presentated at the SPE Reservoir
Simulation Symposium held in Houston, Texas, USA, 23–25 February.
J.J. Arps. Analysis of decline curves [J]. Trans. AIME, 160, 1945, 228–247.
Martín Salazar, Hector Gonzalez, Sébastien Matringe, et al. 2012. Combining Decline-Curve Analysis and Capacitance-
Resistance Models To Understand and Predict the Behavior of a Mature Naturally Fractured Carbonate Reservoir
Under Gas Injection. Paper 153252 presentated at the SPE Latin American and Caribbean Petroleum Engineering
Conference held in Mexico City, Mexico, 16–18 April.
P.H. Gentil. The Use of Multilinear Regression Models in Patterned Waterfloods: Physical Meaning of the Regression
Coefficients, M.S. Thesis, The University of Texas at Austin, 2005.
Xu Shaohua, Bi Congcong, Zhang Yu, et al. Oilfield Development Indicators Prediction Based on Radial Basis Process
Neural Network [J]. Computing Technology and Automation, 2015, 35 (3):52–54.
Zhang Dongmei, Lin Zihang, Kang Zhijiang, et al. Predictionof Development Dynamic Indexs for Fractured-Vuggy
Carbonate Reservoir Based on EEMD and Gaussian Process Autoregression Model [J]. Geological Science and
TechnologyInformation, 2019, 38 (3):256–263.
Zhang Jinqing. A further theoretical discussion on water flooding curve and improvement of Tong's chart [J]. China
Offshore Oil and Gas, 2019, 31 (1):90–97.
Zhao Ling, Li Xuegui, Xu Shaohua, et al. A Model of Oilfield Development Index Forecast Base on Process Support
Vector Regression Machine [J]. Mathematics in Practice and Theory, 2018, 48 (10):83–88.

You might also like