Professional Documents
Culture Documents
Reservoir Porosity and Permeability Estimation from Well Logs using Fuzzy Logic and
Neural Networks
Jong-Se Lim, SPE, Korea Maritime Univ., and Jungwhan Kim, SPE, KODECO Energy Co., Ltd.
the noise in the dataset. The significant inputs to the neural consisting of a number of input signals paired with target
network are identified using fuzzy curves that can identify signals. The inputs are presented to the network and the
relationships between an available parameter and variables in corresponding outputs are calculated with the aim of
noisy data sets1. minimizing the model error, which is the total difference
Consider a data pair (x, y) where x is the event and y is the between calculated outputs and target signals. The back
reactions. The problem is to predict y when x changes slightly, propagation algorithm utilizing the gradient descent method is
in a neighborhood close to x. The fuzzy membership function the most commonly used method to reduce model error. The
of (x, y) gives a local prediction of y according to the training process creates a set of parameters that can be used
information from only (x, y). The fuzzification of the data is for predicting property values in situations where the actual
done with Gaussian function. Fuzzy membership function is output is unknown.
defined as Eq. (1). A typical back propagation neural network (BPNN)
contains of three layers: input, hidden, and output layers. Each
layer is made of a number of processing elements or neurons.
xi − x 2 Each neuron is connected to every neuron in the preceding
Fi ( x) = exp − ⋅ yi (1) layer by a simple weighted link. BPNN requires the use of
b training patterns, and involves a forward propagation step
followed by a backward propagation step. The forward
propagation step sends input signals through the neurons at
Where b defines the shape of the fuzzy membership curves
each layer resulting in an output value. BPNN uses the
and is about 10% of data set range. A fuzzy curve function is
following mathematical function4,5:
used to rank noisy data. The fuzzy curve function gives a
global prediction y because it consists of the sum of the local
predictions (fuzzy membership functions). Fuzzy curve n2
n1
function is defined as Eq. (2). y = f w0 + ∑ w j f j v0 j + ∑ vij xi (3)
j =1 i =1
n
Ci = n1 n 2
(4)
Neural Networks
Neural networks have been successfully used in a variety of ∑∑ w
k =1 j =1
kj
related petroleum engineering applications such as reservoir
characterization, optimal design of stimulation treatments, and
optimization of field operations2,3. The fundamental Where Ci is the average contribution of input variable i, wij is
processing element of a neural network is a neuron. Basically, the connection weight from input neuron i to hidden neuron j.
a biological neuron receives inputs from other sources, This intelligent computing technique can help engineers in
combines them in some way, performs a generally nonlinear solving problems which have not been solved by traditional
operation on the result, and then outputs the result. A typical and conventional computing methods. Neural networks do not
neuron contains a cell body, dendrites, and an axon2. Fig. 3 is require the specification of a structural relationship between
a schematic diagram of a biological neuron. An artificial the inputs and outputs unlike statistical regression analysis.
neural network is a computer model that attempts to mimic Neural networks are used as a nonlinear regression tool to
simple biological learning processes and simulate specific develop transformation between well logs and core analysis
functions of natural neurons in human nervous system. It data. Such a transformation can be used for estimating
learns from examples or experiences, and is extremely useful porosity and permeability in un-cored intervals or wells.
in solving pattern classification and mapping problem. The Recent comparison studies have shown that BPNN models
training or learning phase is an essential starting point for use
of neural networks. This process requires training patterns
SPE 88476 3
may be more accurate than conventional methods and coefficients for the permeability by regression and neural
statistical regression for reservoir properties estimation2,4,5-8. network models compared with measured core data were
0.5654 and 0.9998, respectively (Fig. 12). Fig. 13 shows the
Application estimated permeability and core measured permeability versus
The intelligent technique using fuzzy logic and neural network depth. Multiple regression under-estimates higher permeability
is demonstrated with an application to the well data of Well A, values while neural network shows better consistency in
Block K in offshore Korea. 13.25m of core was recovered and following the actual trend in permeability variation. It was
47 core porosity and permeability values were measured in shown in these results that neural network performs better than
this well. Fig. 5 shows the histogram and descriptive statistics multiple regression method in estimating reservoir porosity
for core measurements. The following 8 conventional well and permeability from well logs.
logs were considered for analysis: neutron log (NPHI), sonic
log (DT), gamma ray log (GR), caliper log (CAL), laterolog Conclusions
deep (LLD), laterolog shallow (LLS), density log (RHOB), In this study, the intelligent technique is used to estimate
and spontaneous potential log (SP). The well log data are reservoir porosity and permeability from conventional well
shown in Fig. 6. logs. Fuzzy curve analysis based on fuzzy logics can be used
The first step is to determine the strength of relationships for selecting the best related parameters with reservoir
between the variables for selecting the best related well logs properties. Excellent correlation coefficients have been
with core porosity and permeability data. We constructed the obtained for porosity and permeability using neural network
cross plots between well logs and core measurements, but models.
found weak correlation based on correlation coefficients and The techniques using fuzzy logic and neural network can
visual observations. Next, fuzzy curve analysis based on fuzzy make more accurate and reliable reservoir properties
logic was utilized to analysis correlations between the estimation compared with conventional methods. This
variables. Normalized data by the maximum-minimum intelligent technique can be utilized a powerful tool for
normalization equation were used for fuzzy curves generation. reservoir characterization from well logs in petroleum
Fig. 7 shows the fuzzy ranked porosity and permeability industry.
curves for each well log. These fuzzy curves could identify
visual relationships between core measurements and well logs References
from noisy data sets. Fuzzy curve analysis could help to select 1. Weiss, W.W., Weiss, J.W., and Weber, J.: “Data Mining at a
the best related well logs with core analysis data as inputs for Regulatory Agency to Forecast Waterflood Recovery,” paper
regressions and neural networks. The ranges of fuzzy ranked SPE 71057 presented at the 2001 SPE Rocky Mountain
curves were used as the ranking criteria. The results of Petroleum Technology Conference held in Keystone, Colorado,
21-23 May.
analyzing porosity and permeability fuzzy curves are tabulated 2. Mohaghegh, S.: “Virtual-Intelligence Applications in Petroleum
in Table 1 and Table 2, respectively. We selected six well logs Engineering: Part I -Artificial Neural Networks,” Journal of
(NPHI, CAL, LLD, LLS, RHOB, and SP) for porosity Petroleum Technology (Sep. 2000) 52, 64-73.
estimation. NPHI, DT, GR, LLD, RHOB, and SP were chosen 3. Tamhane, D., Wong, P.M., Aminzadeh, F., and Nikravesh, M.
for permeability model. “Soft Computing for Intelligent Reservoir Characterization,”
For a comparative study, both multiple variable regressions paper SPE 59397 presented at the 2000 SPE Asia Pacific
and neural networks were applied to the selected well log data Conference on Integrated Modelling for Asset Management held
and the computed results were compared with core measured in Japan, 25-26 April.
porosity and permeability. The neural networks were trained 4. Wong, P.M., Henderson, D.J., and Brooks, L.J.: “Reservoir
Permeability Determination from Well Log Data using Artificial
by a training set with six well logs and core analysis data. Neural Networks: An Example from the Ravva Field, Offshore
Using the same data, we developed the porosity and India,” paper SPE 38034 presented at the 1997 SPE Asia Pacific
permeability models by multiple variable regressions. The Oil and Gas Conference held in Kuala Lumpur, Malaysia, 14-16
ability of a regression model to predict the property extremes April.
is enhanced through a weighting scheme of the high and low 5. Lim, Jong-Se: “Reservoir Permeability Determination using
values. But because of this, the predictor can become unstable Artificial Neural Network,” Journal of the Korean Society for
and also statistically biased. Geosystem Engineering (2003) 40, 232-238.
The contribution of each log for porosity model is shown 6. Balan, B., Mohaghegh, S., and Ameri, S.: “State-Of-The-Art in
in Fig. 8. NPHI contributed the most to the neural network, Permeability Determination From Well Log Data: Part 1-A
Comparative Study, Model Development,” paper SPE 30978
while LLD contributed the least amount. Multiple regressions’ presented at the 1995 SPE Eastern Regional Conference &
correlation coefficient of porosity in Fig. 9 is 0.7640 while Exhibition held in Morgantown, West Virginia, 17-21 Sep.
neural network has a correlation coefficient of 0.9993. 7. Soto B., R., Ardila, J.F., Ferneynes, H., and Bejarano, A.: “Use
Estimated reservoir porosity results from well logs by each of Neural Networks to Predict the Permeability and Porosity of
method are shown in Fig. 10. Fig. 11 presents the computed Zone “C” of the Cantagallo Field in Colombia,” paper SPE
porosity and core porosity versus depth. The regression model 38134 presented at the 1997 SPE Petroleum Computer
gives the best results on the average while neural network Conference held in Dallas, TX, 8-11 June.
provided more accurate results compared with multiple 8. Malki, H.A, Baldwin, J.L., and Kwari, M.A.: “Estimating
regressions. Fig. 11 shows the average contribution of each Permeability by Use of Neural Networks in Thinly Bedded
Shaly Gas Sands,” SPE Computer Application (April 1996) 8,
well log data to neural network for permeability model. DT 58-62.
was the most contributed log to the network. The correlation
4 SPE 88476
(a)
(b)
(b)
(a)
δZ δY E
d
α α
X X
∆V ∆W
(a)
(a) (b)
(b)
1.0
1.0 1.0
1.0
0.8
0.8 0.8
0.8
Value
Value
RankedValue
RankedValue
0.6
0.6 0.6
0.6
FuzzyRanked
FuzzyRanked
0.4
0.4 0.4
0.4
Fuzzy
Fuzzy
0.2
0.2 0.2
0.2
porosity
porosity porosity
porosity
permeability
permeability permeability
permeability
0.0
0.0 0.0
0.0
0.0
0.0 0.2
0.2 0.4
0.4 0.6
0.6 0.8
0.8 1.0
1.0 0.0
0.0 0.2
0.2 0.4
0.4 0.6
0.6 0.8
0.8 1.0
1.0
Neutron
NeutronLog
Log Sonic
SonicLog
Log
(c)
(c) (d)
(d)
1.0
1.0 1.0
1.0
0.8
0.8 0.8
0.8
Value
Value
RankedValue
RankedValue
0.6
0.6 0.6
0.6
FuzzyRanked
FuzzyRanked
0.4
0.4 0.4
0.4
Fuzzy
Fuzzy
0.2
0.2 0.2
0.2
porosity
porosity porosity
porosity
permeability
permeability permeability
permeability
0.0
0.0 0.0
0.0
0.0
0.0 0.2
0.2 0.4
0.4 0.6
0.6 0.8
0.8 1.0
1.0 0.0
0.0 0.2
0.2 0.4
0.4 0.6
0.6 0.8
0.8 1.0
1.0
Gamma
GammaRay
RayLog
Log Caliper
CaliperLog
Log
(e)
(e) (f)
(f)
1.0
1.0 1.0
1.0
0.8
0.8 0.8
0.8
Value
Value
RankedValue
RankedValue
0.6
0.6 0.6
0.6
FuzzyRanked
FuzzyRanked
0.4
0.4 0.4
0.4
Fuzzy
Fuzzy
0.2
0.2 0.2
0.2
porosity
porosity porosity
porosity
permeability
permeability permeability
permeability
0.0
0.0 0.0
0.0
0.0
0.0 0.2
0.2 0.4
0.4 0.6
0.6 0.8
0.8 1.0
1.0 0.0
0.0 0.2
0.2 0.4
0.4 0.6
0.6 0.8
0.8 1.0
1.0
LaterologDeep
Laterolog Deep LaterologShallow
Laterolog Shallow
(g)
(g) (h)
(h)
1.0
1.0 1.0
1.0
0.8
0.8 0.8
0.8
Value
Value
RankedValue
RankedValue
0.6
0.6 0.6
0.6
FuzzyRanked
FuzzyRanked
0.4
0.4 0.4
0.4
Fuzzy
Fuzzy
0.2
0.2 0.2
0.2
porosity
porosity porosity
porosity
permeability
permeability permeability
permeability
0.0
0.0 0.0
0.0
0.0
0.0 0.2
0.2 0.4
0.4 0.6
0.6 0.8
0.8 1.0
1.0 0.0
0.0 0.2
0.2 0.4
0.4 0.6
0.6 0.8
0.8 1.0
1.0
Density
DensityLog SP
Log SPLog
Log
Fig. 7-Fuzzy ranked porosity and permeability curves for well logs
8 SPE 88476
SP log
11.9% Neutron log
25.4%
Density log
22.2%
Caliper log
15.1%
Laterolog
Laterolog Deep
Shallow
11.8%
13.5%
Fig. 8-Average contribution of each input well log data to neural network for porosity model
(a)
(a) (b)
(b)
0.20 0.20
0.20 0.20
Correlation coefficient = 0.7640 Correlation coefficient = 0.9993
Correlation coefficient = 0.7640 Correlation coefficient = 0.9993
0.15 0.15
0.15 0.15
Estimated porosity
Estimated porosity
Estimated porosity
Estimated porosity
0.10 0.10
0.10 0.10
0.05 0.05
0.05 0.05
0.00 0.00
0.00 0.00
0.00 0.05 0.10 0.15 0.20 0.00 0.05 0.10 0.15 0.20
0.00 0.05 0.10 0.15 0.20 0.00 0.05 0.10 0.15 0.20
Meas ured poros ity Meas ured poros ity
Meas ured poros ity Meas ured poros ity
Fig. 9-Cross plots of core porosity and estimated porosity (a) by multiple regressions (b) by neural network
(a) Porosity
Porosity (b) Porosity
Porosity
0.0 0.1 0.2 0.3 0.0 0.1 0.2 0.3
0.0 0.1 0.2 0.3 0.0 0.1 0.2 0.3
2347 2347
2347 2347
2349 2349
2349 2349
2351 2351
2351 2351
2353 2353
2353 2353
Depth(m)
Depth(m)
Depth(m)
Depth(m)
2355 2355
2355 2355
2357 2357
2357 2357
2359 2359
2359 2359
2361 2361
2361 2361
2363 2363
2363 2363
Core poros ity Es timated poros ity Core poros ity Es timated poros ity
Core poros ity Es timated poros ity Core poros ity Es timated poros ity
Fig. 10-Estimated reservoir porosity from well logs (a) by multiple regressions (b) by neural network
SPE 88476 9
Sonic log
24.2%
Laterolog Deep
16.1%
(a)
(a) (b)
(b)
100 100
100 100
Correlation coefficient = 0.5654 Correlation coefficient = 0.9998
Correlation coefficient = 0.5654 Correlation coefficient = 0.9998
75 75
Estimated permeability(md)
Estimated permeability(md)
75 75
Estimated permeability(md)
Estimated permeability(md)
50 50
50 50
25 25
25 25
0 0
0 0
0 25 50 75 100 0 25 50 75 100
0 25 50 75 100 0 25 50 75 100
Meas ured permeability (md) Meas ured permeability (md)
Meas ured permeability (md) Meas ured permeability (md)
Fig. 12-Cross plots of core permeability and estimated permeability (a) by multiple regressions (b) by neural network
(a) Permeability(md)
Permeability(md) (b) Permeability(md)
Permeability(md)
0 40 80 120 0 40 80 120
0 40 80 120 0 40 80 120
2347 2347
2347 2347
2349 2349
2349 2349
2351 2351
2351 2351
2353 2353
2353 2353
Depth(m)
Depth(m)
Depth(m)
Depth(m)
2355 2355
2355 2355
2357 2357
2357 2357
2359 2359
2359 2359
2361 2361
2361 2361
2363 2363
2363 2363
Core permeability Es timated permeability Core permeability Es timated permeability
Core permeability Es timated permeability Core permeability Es timated permeability
Fig. 13-Estimated reservoir permeability from well logs (a) by multiple regressions (b) by neural network