Vous êtes sur la page 1sur 3

International Conference on Computer Science and Service System (CSSS 2014)

Oil production predicting with modified BP neural network method

Haohan Liu, Wei Li, Songlin Zhang

Sichuan College of Architectural Technology
Deyang, China, 618000

AbstractFeasibility of oil production predicting results layer BP ANN is used to establish the ANN model with
influence the annual planning and long-term field development prediction function. The first layer of BP-ANN is input
plan of oil field, so the selection of predicting models plays a layer, the second layer of it is middle layer, and the third
core role. In this paper, a common and useful model is layer of it is the output layer, see figure 1.
introduced, it is,the neural network model. By using this
model to predict the oil production in DAQ oilfield in China, Midd
advantages and disadvantages of the model has been discussed. g1 le h1
The predicting results show: the fitting accuracy by the neural Input Outp
network model is high, and the prediction error is smaller ut
U1 g2 h2 X1
than 10%, so neural network model can be used to short-term
forecast of oil production, after changing the weighting value U2 X2
in training, we can also improve the predicting accuracy,
however, this process takes much time. Next, our team will try
to develop new theory to shorten the training time.

Keywords-Oil field;oil production; neural network; U Xn

predicting accuracy gp- hp-
1 1
gp hp
Oil production prediction is very important in the oilfield
Figure1 BP neural network structure
development, study on oil production predicting method is a
key topic of petroleum science. At present, there are four Network simulation
kinds of predicting methods[1-6] with physical meanings: Assume N -samples (input-output data samples) are
Empirical formula such as Arps method, Hubbert model, used during the training process:
water-flood decline curve method; Hydrodynamic model The input variables are
U t u1 t , u2 t ,, um t ,the output variables are
based on fluid mechanics model; Material balance equation
model; Numerical reservoir simulation model. Besides the
X t x1 t , x2 t ,, xn t . For the k -sample
above mentioned four methods, there is a typical type of
prediction model related to modern optimization, this model
type is composed of GM(1,1)model, neural network model, (k 1, 2 N ) ,let U k (u1k , u2 k ,us k ) be input mode
support vector machine model, etc. Oil field development
vector; X k ( x1k , x2k xnk ) be the expectation output
system is a complex multi variables non-linear dynamical
systems, different predicting model has different vector; G k ( g1k , g2 k , g p k ) be the middle layer input
characteristics like predicting accuracy.Neural network
model is an effective method to solve multi nonlinear vector; H k (h1k , h2 k , hp k ) be the output unit vector of
mapping problem. At present it is used in many disciplines, middle layer; M k (m1k , m2 k , mn k ) be the input vector
even in the oil production prediction. In this paper, neural
of output layer; X ( x1 , x2 xn ) be the output vector of
k k k k
network model is used to predict the oil production of
XINJIANG oil field, and good predicting results have been
achieved.. output layer; {Wij , i 1, 2 m; j 1, 2 p} be the
connection weights from input layer to middle
layer; {V jt , j 1, 2 p; l 1, 2 n} be the connection
At present, in the application of artificial neural networks
weights from middle layer to output
(ANN) [7, 8], most of them are back propagation (BP) ANN
and their variations. It has been proved that BP neural layer; { j , j 1, 2 p} be the threshold value of middle
network can approximate any multivariate continuous layer; { l , l 1, 2 n} be the threshold value of output
function. Kolmogorov rule guaranteed that any continuous
function or mapping can be achieved by a 3-layer ANN. 3- layer; (0 1) is the learning rate.

2014. The authors - Published by Atlantis Press 146

Let the response function of ANN f ( x ) be Sigmoid- III. EASE OF USE OF APPLICATIONS AND DISCUSSIONS
x 1
type function: f ( x) (1 e ) Given the initial oil production data (from 1958 to 2012) of
certain oilfield block in China, then the above-mentioned
The input and output values of each neural unit satisfy three method can be used to predict the oil production of
the following relationship: different oilfield block(A1,A2,A3). After using the BP
Middle layer: neural network predicting model gives figure2, figure3 and

intput:g j wij ui j (i 1, 2 m; j 1, 2 p)
k k
(1) figure4:
i 1
output:h k f ( g k )(k 1, 2 N )

j j

Output layer:

input : ml v jl h j l (l 1, 2n)
k k 10

j 1 (2)
output:x k f (mk )
l l

After training of N samples, the network error is:

1 N n
( xlk xlk )2
0 10 20 30 40 50 60
Etotal .
2 k 1 l 1 Figure2 oil production prediction value with BP neural
Error of output layer unit: network method in block-A1
dlk ( xlk xlk ) xlk (1 xlk ) (3) 2.5

Error of middle layer unit:

ekj [ dl k v jl ]h j k (1 h j k )

l 1 1.5

The connection weights V

jl and the threshold

value { t } can
be modified by the output layer
error {dl } and output value of middle layer unit h j : k 0.5

V jl (q 1) V jl (q) dlk h kj
(5) 0 5 10 15 20 25 30 35 40 45 50

k 1
Figure3 oil production prediction value with BP neural
l (q 1) l (q) d l
network method in block-A2
k 1 7

The connection weights {Wij } and the threshold 6

value { j } can be modified by the middle layer


error {e kj } and input value of input


layer U (u1 , u2 ,um ) :

k k k k

Wij (q 1) Wij (q) ekj uik (7)
k 1
j (q 1) j (q) ekj
(8) 0 10 20 30 40 50 60

k 1
Figure4 oil production prediction value with BP neural
Repeat the above-mentioned learning mode, until the network
network method in block-A3
converges to a given error range.
In figure2, figure3and figure4, the black circle means
the real oil production, the colored curve is the predicting

curve with ANN model. ANN model can be used to short [2] Hu Jianguo, Chen Yuanqian, Zhang Shengzong. A new
term prediction between 15 to 40 in figure 2to 4. model using for prediction of oil-gas field production, ACTA
PETROLEI SINICA, Vol.16, No.1, 1995, pp.79-86.
[3] Chen Gan,Yang Xinling. AN EXAMINATION INTO
1Prediction with ANN model model can comply with the AND GAS(GEOLOGY), Vol.16, No.4, 2002, pp. 254-259.
actual oilfield production dynamics, the prediction errors of [4] Chen Yuanqian, Zhao Qingfei. A Combined Method of
Weibull Model and Water Drive Curve of B Type,
them are less than 10%. XINJIANG PETROLEUM GEOLOGY, 2000, 21(5): 405-407.
2It is a learning type of model; much data is needed to [5] Chen Yuanqian, Hu Jianguo, Zhang Dongjie. Deduction of
complete the prediction, so they are only suitable for the Logistic model and auto regression methods, XINJIANG
short term prediction. PETROLEUM GEOLOGY, Vol.17, No.2, 1996, pp. 150-155.
3We can improve the predicting accuracy by modifying [6] Yue Ping, Chen Xiaofan, Cui Ligong, etc. suggestion
solution of Wengs model and Logistic model in production
the weighting valuer. research, Journal of oil and gas technology, Vol.31, No.4,
2009, pp. 277-279.
[7] Yang Bin. Applied neural network in oil well
REFERENCES logging,Beijing: Petroleum Industry Press,2005
[8] Wu Xingen, Ge Jiali. oil production forecast with ANN
method,Petroleum Exploration and Development, Vol.21,
No.3, 1994, pp.31-42.
[1] Ji Bingyu. Overview of Oilfield Development Indicators
Forecasting Methods, Petroleum geology and oilfield
development in DAQING, Vol.18, No.2, 1999, pp.19-22.