Research ArticleGeoelectrical Data Inversion by Clustering Techniques ofFuzzy Logic to Estimate the Subsurface Layer Model
A. Stanley Raj,1 D. Hudson Oliver,2 and Y. Srinivas3
1Department of Physics, Vel Tech University, Avadi, Chennai 600062, India2Department of Physics, Senthamarai College of Arts and Science, Palkalai Nagar, Madurai 625021, India3Centre for Geotechnology, Manonmaniam Sundaranar University, Tirunelveli, Tamil Nadu 627012, India
Correspondence should be addressed to A. Stanley Raj; stanleyraj [email protected]
Received 25 July 2014; Accepted 19 January 2015
Academic Editor: Robert Tenzer
Copyright Β© 2015 A. Stanley Raj et al. This is an open access article distributed under the Creative Commons Attribution License,which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.
Soft computing based geoelectrical data inversion differs from conventional computing in fixing the uncertainty problems. It istractable, robust, efficient, and inexpensive. In this paper, fuzzy logic clustering methods are used in the inversion of geoelectricalresistivity data. In order to characterize the subsurface features of the earth one should rely on the true field oriented data validation.This paper supports the field data obtained from the published results and also plays a crucial role in making an interdisciplinaryapproach to solve complex problems.Three clustering algorithms of fuzzy logic, namely, fuzzy πΆ-means clustering, fuzzyπΎ-meansclustering, and fuzzy subtractive clustering, were analyzed with the help of fuzzy inference system (FIS) training on syntheticdata. Here in this approach, graphical user interface (GUI) was developed with the integration of three algorithms and the inputdata (AB/2 and apparent resistivity), while importing will process each algorithm and interpret the layer model parameters (trueresistivity and depth). A complete overview on the three above said algorithms is presented in the text. It is understood from theresults that fuzzy logic subtractive clustering algorithm gives more reliable results and shows efficacy of soft computing tools in theinversion of geoelectrical resistivity data.
1. Introduction
In recent years, soft computing was bound to play a keyrole in the earth sciences. This is in part due to the subjectnature of the rules governing many physical phenomena inthe earth sciences. As our problems related to nonlinearparameters of earth, it becomes too complex to rely onlyon one discipline and we find ourselves at the midst ofinformation explosion interdisciplinary analysis methods.To solve complex problems, we need to rely on knowledgebased approach than standard mathematical techniques.Instead, we need to complement the conventional analysismethods with a number of emerging methodologies andsoft computing techniques such as expert systems, artificialintelligence, neural network, fuzzy logic, genetic algorithm,probabilistic reasoning, and parallel processing techniques.
Many of the researchers [1β13] addressed the problems ininterdisciplinary approach.
Recent applications of soft computing techniques havealready begun to enhance our ability of estimating thesubsurface features of earth. In this paper vertical electricalsounding (VES) data obtained from the field is fed as an inputto the FIS training, where it generates many synthetic datanecessary for clustering algorithm. After getting the syntheticdata,MATLABbased program runs on the specially designedmajor algorithm (Figure 2). The data processing depends onvarious parameters, mainly the number of iterations (user-dependent) and error percentage. The lowest error percentthus provides the best performance of output (true resistivityand depth) information of subsurface earth.
2. Geophysical Method
Schlumberger electrode array is used to study the elec-trical resistivity distribution of the subsurface in order tounderstand the groundwater conditions such as resistivity,
Hindawi Publishing CorporationInternational Journal of GeophysicsVolume 2015, Article ID 134834, 11 pageshttp://dx.doi.org/10.1155/2015/134834
2 International Journal of Geophysics
BatteryI
V
A M N B
Current flow through earth
Figure 1: Schlumberger electrode configuration.
thickness, and depth (Figure 1). Usually the depth of penetra-tion is proportional to the separation between the electrodesand varying the electrode separation provides informationabout the stratification of the ground.
The apparent resistivity value depends on the electricalconductivities of different rocks andminerals.Thus electricalprospecting can be carried out to understand the subsurfaceearth. The data collected from the field has been interpretedusing fuzzy clustering algorithms.TheFIS algorithmprovidesthe necessary database needed for interpretation. Moreover,the best model of the trained database fits with the apparentresistivity curve. The corresponding layer model will beproduced as an output with lowest root mean square errorin particular number of iterations.
3. Fuzzy Logic Applications onGeoscience Data Inversion
Fuzzy logic is considered to be appropriate to deal with thenature of uncertainty in system and human errors, which arenot included in current reliability theories. The basic theoryof fuzzy sets was first introduced by Zadeh [12].
In recent years, fuzzy logic, or more generally, fuzzy settheory, has been applied extensively in many geophysicalcharacterization studies. Fuzzy set theory has the abilityto deal with such information and to combine it withthe quantitative observations. The applications are many,including resistivity inversion, magnetic studies, seismic andstratigraphic modeling, and formation evaluation.
Nordlund [14] has presented a study on dynamic strati-graphic modeling using fuzzy logic. Cuddy [15] has appliedfuzzy logic to solve a number of petrophysical problems inseveral North Sea fields. Fang and Chen [16] also appliedfuzzy rules to predict porosity and permeability from five
compositional and textural characteristics of sandstone inthe Yacheng Field (South China Sea). Huang et al. [17]have presented a simple but practical fuzzy interpolator forpredicting permeability from well logs in the North WestShelf (offshore Australia).The basic idea was to simulate localfuzzy reasoning [18, 19] proposed by Bois in applying the useof fuzzy sets theory in the interpretation of seismic sections.
3.1. Fuzzy Inference System (FIS) Training. The fuzzy infer-ence system is a popular computing framework based onthe concepts of fuzzy set theory, fuzzy if-then rules, andfuzzy reasoning. It has found successful applications in a widevariety of fields, such as automatic control, data classification,decision analysis, expert systems, time series prediction,robotics, and pattern recognition.
The implementation process of nonlinear mappingbetween AB/2 values and apparent resistivity data values bymeans FIS is shown in primary class training of Figure 2(a).This mapping is accomplished by a number of fuzzy if-thenrules, each of which describes the local behaviour of themapping. In particular, the antecedent of a rule defines a fuzzyregion in the input space where the imported data stored willconsequently specify the output in the fuzzy region.
4. Clustering Tool to Invert Geoelectrical Data
The idea of data grouping, or clustering, is simple in its natureand is close to the human way of thinking; whenever we arepresented with a large amount of data, we usually tend tosummarize this huge number of data into a small number ofgroups or categories in order to further facilitate its analysis.Moreover, most of the data collected in many nonlinearproblems related to earth seem to have some inherent proper-ties that lend themselves to natural groupings. Nevertheless,
International Journal of Geophysics 3
Inverting the synthetic data using slope variation method
(true resistivity and depth)]
Output multilayer model
Regressed output layer model (true resistivity and depth)
Synthetic data generated after the completion of FIS training
Applying the clustering algorithm of fuzzy logic to classify the best fit model
compress the output multilayer modelApplying linear regression technique to
Output = [slope variation layer model for corresponding synthetic dataInput = [synthetic data]
(a)
(b)
Importing AB/2 and apparentresistivity data (πa)
Initialize fuzzy inference system (FIS) training with input-AB/2 and output-apparent resistivity (πa)
Figure 2: Flowchart showing the specially designed fuzzy clustering algorithm with (a) primary class training and (b) major class training.
finding these groupings or trying to categorize the data is nota simple task for humans. This is why some methods in softcomputing-clustering technique have been proposed to solvegeoelectrical resistivity inversion problem.
In clustering, three algorithms were employed, namely,fuzzy πΆ-means clustering, fuzzy πΎ-means clustering, andfuzzy subtractive clustering. For clustering, a large numberof synthetic databases had been developed during the FIStrainingwhichwill uniquely play a role in data interpretation.Figure 2(b) thus provides the major class training for theinversion of synthetic data to layer model based on thefollowing clustering algorithms.
4.1. Fuzzy πΆ-Means Clustering. Fuzzy πΆ-means clustering(FCM) is a data clustering algorithm inwhich each data pointbelongs to a cluster to a degree specified by a membershipgrade. Bezdek in 1981 [20] proposed this algorithm as animprovement over hard πΆ-means clustering algorithm. Eachof the π data pairs belongs to each of the π groups with amembership coefficient, π’
ππ, being the membership degree of
pair π to cluster π. Letπ·2ππ be the distance between pair π andcluster π, basically defined as the Euclidean norm and moregenerally as
π·2ππ =
π₯π β Vπ
2
π΄= (π₯πβ Vπ) π΄ (π₯
πβ Vπ)π, (1)
4 International Journal of Geophysics
π₯πbeing the πth data pair used for the clustering, π΄ being a
positive definite symmetricmatrix, and Vπbeing the prototype
of cluster π.FCM partitions a collection of π vectors obtained from
the imported field data π₯π, π = 1, . . . , π, into π fuzzy groups and
finds a cluster center in each group such that a cost functionof dissimilarity measure is minimized. To accommodate theintroduction of fuzzy partitioning, the membership matrixπ is allowed to have elements with values between 0 and 1.The normalised values of AB/2 and apparent resistivity datathus plotted with the cluster centres were shown in the GUIpanel of Figures 4 and 8 for the corresponding data. However,imposing normalization stipulates that the summation ofdegrees of belongingness for a data set always be equal tounity:
π
β
π=1
π’ππ= 1, βπ = 1, . . . , π. (2)
The cost function (or objective function) for FCM is then ageneralization of (1):
π½ (π, π1, . . . , π
π) =
π
β
π=1
π½π=
π
β
π=1
π
β
π=1
π’π
πππ2
ππ, (3)
where π’ππis between 0 and 1; π
πis the cluster center of fuzzy
group π; πππ= βππβ π₯πβ is the Euclidean distance between
πth cluster center and πth data point; and π β [1,β) is aweighting exponent.
The necessary conditions for (3) to reach a minimum canbe found by forming a new objective function π½ as follows:
π½ (π, π1, . . . , π
π, π1, . . . , π
π)
= π½ (π, π1, . . . , π
π) +
π
β
π=1
ππ(
π
β
π=1
π’ππβ 1)
=
π
β
π=1
π
β
π=1
π’π
πππ2
ππ+
π
β
π=1
ππ(
π
β
π=1
π’ππβ 1) ,
(4)
where ππ, π = 1 to π, are Lagrange multipliers for the π con-
straints in (2). By differentiating π½ (π, π1, . . . , π
π, π1, . . . , π
π)
with respect to all its input arguments, the necessary condi-tions for (3) to reach its minimum are
ππ=
βπ
π=1π’π
πππ₯π
βπ
π=1π’π
ππ
, (5)
π’ππ=
1
βπ
π=1(πππ/πππ)2/(πβ1)
. (6)
The fuzzyπΆ-means algorithm is simply an iterated procedurethrough the preceding two necessary conditions.
Step 1. Initialize the membership matrix π with randomvalues between 0 and 1 taken with the constraint of not beinglosing the essence of the field apparent resistivity data bysatisfying (2).
Step 2. Calculate π fuzzy cluster centers ππ, π = 1, . . . , π, using
(5).
Step 3. Compute the cost function according to (3). Stop ifeither it is below a certain tolerance value or its improvementover previous iteration is below a certain threshold.
Step 4. Compute a new π using (6). Go to Step 2 and repeatthe steps until the performance goal is reached.
The cluster centers can also be first initialized and thenthe iterative procedure can be carried out. The performancedepends on the initial cluster centers, thereby allowing useither to use another fast algorithm to determine the initialcluster centers or to run FCM several times, each startingwith a different set of initial cluster centers. Therefore, ateach iteration, different cluster centers will be formed andthe centers which focus finally will fall on the exact syntheticdata that correlates with the field data with minimum errorpercentage.
4.2. Fuzzy πΎ-Means Clustering. The πΎ-means clustering [21,22] has been applied to a variety of areas, including image andspeech data compression [23], data preprocessing for systemmodelling [24], and task decomposition [23].
The πΎ-means algorithm partitions a collection of πvectors which is the input data AB/2 and apparent resistivitydata π₯
π, π = 1, . . . ., π, into π groups πΊ
π, π = 1, . . . , π, and finds
a cluster center in each group such that a cost function (oran objection function) of dissimilarity (or distance) measureis minimized. When the Euclidean distance is chosen as thedissimilarity measure between a vector π₯
πin group π and the
corresponding cluster ππ, the cost function can be defined by
π½ =
πΆ
β
π=1
π½π=
πΆ
β
πΌ=1
( β
π,π₯πβπΊπ
π₯π β ππ
2) , (7)
where π½π= βπ,π₯πβπΊπ
βπ₯πβππβ2 is the cost functionwithin group
π. Thus, the value of π½πdepends on the geometrical properties
of πΊπand the location of π
π.
In general, a generic distance function π(π₯π, ππ) can be
applied for vector π₯πin group π; the corresponding overall
cost function is thus expressed as
π½ =
πΆ
β
πΌ=1
π½πΌ=
πΆ
β
πΌ=1
( β
π,π₯πβπΊπ
π (π₯πβ ππ)) . (8)
For simplicity, the Euclidean distance is used as the dissimi-larity measure and the overall cost function is expressed as in(7).
The partitioned groups are typically defined by a π Γ πbinary membership matrixπ, where the element π’
ππis 1 if the
πth data point π₯πbelongs to group π and 0 otherwise. Once
the cluster centers ππare fixed, the minimized π’
ππfor (7) can
be derived as follows:
π’ππ={
{
{
1 if π₯π β ππ
2
β€π₯πβ ππ
2
, for each π ΜΈ= π,
0 otherwise.(9)
International Journal of Geophysics 5
Restating, π₯πbelongs to group π if π
πis the closest center
among all centers. Since a given data point can only bein a group, the membership matrix π has the followingproperties:
π
β
π=1
π’ππ= 1, βπ = 1, . . . , π,
π
β
π=1
π
β
π=1
π’ππ= π.
(10)
On the other hand, if π’ππis fixed, then the optimal center
ππthat minimizes (7) is the mean of all vectors in group π:
ππ=1
πΊπ
β
π,π₯πβπΊπ
π₯π, (11)
where |πΊπ| is the size of πΊ
πor |πΊπ| = βπ
π=1π’ππ.
For a batch-mode operation, the πΎ-means algorithm ispresented with a data set π₯
π, π = 1, . . . , π; the algorithm
determines the cluster centers ππand the membership matrix
π iteratively using the following steps.
Step 1. Initialize the cluster center ππ, π = 1, . . . , π. Here in this
approach it has been randomly chosen within the range ofthe minimum and maximum values of normalised apparentresistivity data.
Step 2. Determine the membership matrix π by (9).
Step 3. Compute the cost function according to (7). Stop ifeither it is below a certain tolerance value or its improvementover previous iteration is below a certain threshold. Iterationscan be fixed by the user by means of GUI panel.
Step 4. Update the cluster centers according to (11). Goto Step 2 and repeat the steps until the minimum errorpercentage is reached.
The algorithm is inherently iterative, and the performanceof the πΎ-means algorithm depends on the initial positionsof the cluster centers, thereby making it advisable either toemploy some front-end methods to find good initial clustercenters or to run the algorithm several times, each with adifferent set of initial cluster centers. So thereby providing thesynthetic data will be smoothened in order to correlate withthe actual field data.
4.3. Fuzzy Subtractive Clustering. The subtractive clusteringtechnique was proposed by [25], in which data points (notgrid points) are considered as the candidates for clustercenters. By using this method, the computation is simplyproportional to the number of resistivity data points andindependent of the dimension of the inverse problem underconsideration.
Consider a collection of π data points {π₯1, . . . , π₯
π} in
an π-dimensional space. Without loss of generality, theapparent resistivity data points are assumed to have beennormalized within a hypercube. Since each data point is
a candidate for cluster centers, density measure at data pointπ₯πis defined as
π·π=
π
β
π=1
exp(βπ₯πβ π₯π
2
(ππ/22)
) , (12)
where ππis a positive constant. Hence, a data point will have
a high density value if it has many neighbouring data points.The radius π
πdefines a neighbourhood; apparent resistivity
data points outside this radius contribute only slightly to thedensity measure.
After the density measure of each data point has beencalculated, the data point with the highest density measure isselected as the first cluster center. Let π₯
π1be the point selected
and π·π1its density measure. Next, the density measure for
each data point π₯πis revised by the formula
π·π= π·πβ π·π1exp(β
π₯πβ π₯π1
2
(ππ/2)2) , (13)
where ππis a positive constant. Therefore, the data points
near the first cluster center π₯π1will have significantly reduced
density measures, thereby making the points unlikely to beselected as the next cluster center. The constant π
πdefines a
neighbourhood that has measureable reductions in densitymeasure.The constant π
πis normally larger than π
πto prevent
closely spaced cluster centers; generally ππis equal to 1.5π
π.
After the density measure for each data point is revised,the next cluster center π₯
π2is selected and all of the density
measures for data points are revised again. This processis repeated until a sufficient number of cluster centers aregenerated.
When applying subtractive clustering to a set of input-output data, each of the cluster centers represents a prototypethat exhibits certain characteristics of the system to bemodelled. These cluster centers would be reasonably used asthe centers for the fuzzy rulesβ premise in a zero-order Sugenofuzzy model. For instance, assume that the center for the πthcluster is π
πin an π dimension. The π
πcan be decomposed
into two component vectors ππand π
π, where π
πis the input
part and it contains the first π element of ππ; ππis the output
part and it contains the lastπ-π elements of ππ. Then, given
an input vector π₯, the degree to which fuzzy rule π is fulfilledis defined by
ππ= exp(β
π₯ β ππ
2
(ππ/22)
) . (14)
After these procedures are completed, more accuracy canbe gained by using gradient descent or other advancedderivative-based optimisation schemes for further refine-ment.
5. Step by Step Procedure
The workflow of the GUI panel works on the path ofalgorithm description shown in the flowchart of Figure 2.
6 International Journal of Geophysics
(i) Importing AB/2 and apparent resistivity data neededfor interpretation can be done by the push buttonshown in Figure 3(a).
(ii) The corresponding data importedwill be shown in thetable panel of Figure 3(c).
(iii) The user optional slider (Figure 3(b)) used to estimatethe number of iterations needed for the clustering toolto run.
(iv) Push button of Figure 3(d)will help the user tomodifythe imported table values. After editing the necessaryvalues, the user can click the push button shownbelow the table of the GUI to import the modifieddata.
(v) Figure 3(e) provides the different clustering algorithmpush buttons where the corresponding program willrun after importing the necessary data.
(vi) The output layer model and cluster panel graph areshown in Figures 3(f) and 3(g), respectively.
(vii) Running message will be shown in the GUI panel ofFigure 3(h).
(viii) After iterating the algorithm, the user can save therespective plots and can exit easily while pushing thecorresponding push buttons.
6. Results and Discussions
For validating the algorithm and comparative analysis, resis-tivity data of different geological regions and the resultantperformance will conclude the result. The performance mea-sures show that subtractive clustering algorithm result movesmore positively than the other two algorithms according tothis application. In subtractive clustering technique, the near-est neighbourhood radius adjustment based on the densitymeasure of each apparent resistivity data point was converg-ing in each and every set of iterations, thereby concluding thegood performance. If the raw field data contains more noisesor field errors, the converging rate will be slow, but it can beachieved by increasing more number of iterations. Moreover,the adjustments in the dimensionality problems have alreadybeen done in each of the clustering algorithms. Therefore,there will be no dimensionality problems occurring whileiterating each algorithm.
Data 1 was chosen from the Singhbhum Shear Zone ofJaduguda, Jharkhand, India [26]. GUI panel of Figure 3 showsthe interpreted model with successful clustering classifica-tions based on centres.
Figures 4(a)β4(c) show the cluster centres of fuzzy πΆ-means, fuzzy πΎ-means, and fuzzy subtractive clusteringalgorithms, respectively.
The performance measure graph is shown in Figure 6.Data 2was chosen fromKamuli district, EasternUganda [27].Fuzzy based inverted results ofData 2 are shown inGUI panelof Figure 7.
Figures 8(a)β8(c) show the cluster centres of fuzzy πΆ-means, fuzzy πΎ-means, and fuzzy subtractive clusteringalgorithms, respectively.
Table 1: Accuracy of the three fuzzy clustering algorithms for data1.
Number ofiterations
AccuracyFuzzyπΆ-meansclusteringalgorithm
FuzzyπΎ-meansclusteringalgorithm
Fuzzy subtractiveclusteringalgorithm
10 91.6558 96.819 97.569220 94.1628 93.8174 97.723330 92.6629 96.1407 97.897840 85.8052 96.8152 97.821550 96.2261 96.73 97.91100 95.551 96.7432 97.976Overallperformance 92.6773 96.17758 97.8163
Table 2: Accuracy of the three fuzzy clustering algorithms for data2.
Number ofiterations
AccuracyFuzzyπΆ-meansclusteringalgorithm
FuzzyπΎ-meansclusteringalgorithm
Fuzzy subtractiveclusteringalgorithm
10 95.7703 93.3317 95.738920 96.1838 95.2137 96.213230 93.3317 94.9628 96.582840 95.5473 95.5139 97.013250 95.6231 96.1325 97.1101100 95.71 96.1435 97.7123Overallperformance 95.36103333 95.21635 96.72841667
Figures 9 and 10 show the output panel and performanceplot for data 2 (see Figure 5 for data 1 output panel). Oneof the major causes for successful interpretation of fuzzysubtractive clustering technique is that, each and every timewhile iterating the algorithm, the cluster centre has beenrevised on the basis of densitymeasure of each resistivity dataproblem. In this geoelectrical-computational approach, it isvery much essential to update the density measures of eachdata point, as an apparent resistivity curve can fit with manymodels while iterating. The algorithm stops running aftergetting the reliable model by limiting the root mean squareerror as much as possible for reducing the computationaltime.
Theperformance of these techniqueswas compared usingthe two different datasets along with their lithologs. Theperformance measures of data 1 and data 2 were tabulatedin Tables 1 and 2, respectively, which proclaims that the sub-tractive clustering inversion technique seems to be the bestalgorithm for the inversion of geoelectrical resistivity datacomparatively. On the other hand, fuzzy πΆ-means clusteringsuffers inconsistency in the performance. Fuzzy πΎ-meansprovides similar favourable accuracy on comparing with thefuzzy subtractive clustering, but it suffers in producing more
International Journal of Geophysics 7
(a) (e)(f)
(g)
(h)
(b)
(c)
(d)
Figure 3: GUI panel showing the fuzzy subtractive clustering inversion for data 1.
1.2
1
100
0.8
0.8
0.6
0.6
0.4
0.4
0.2
0.2
Field dataCluster centersSynthetic data
(a)
1.2
1.4
1
0
0.8
0.6
0.4
0.2
10 0.80.60.40.2
Field data
Cluster centersSynthetic data
(b)
1.2
1.4
1
0
0.8
0.6
0.4
0.2
10 0.80.60.40.2
Field dataCluster centersSynthetic data
(c)
Figure 4: Formation of fuzzy cluster centers while interpreting data 1: (a) fuzzy πΆ-means clustering, (b) fuzzy πΎ-means clustering, and (c)fuzzy subtractive clustering.
8 International Journal of Geophysics
Figure 5: Output panel showing the fuzzy subtractive clustering inverted model for data 1.
8486889092949698
Accu
racy
0 50 100 150Number of iterations
R2 = 0.1205
y = 0.0411x + 90.964
(a)
93
94
95
96
97
98
Accu
racy
0 50 100 150
Number of iterations
R2 = 0.1289
y = 0.0133x + 95.622
(b)
97.5
97.6
97.7
97.8
97.9
98
98.1
0 50 100 150
Accu
racy
Number of iterations
R2 = 0.6402
y = 0.0037x + 97.661
Best fitLinear (best fit)
(c)
Figure 6: Performance measurement of data 1 for the three algorithms: (a) fuzzy πΆ-means clustering, (b) fuzzy πΎ-means clustering, and (c)fuzzy subtractive clustering.
Figure 7: GUI panel showing the fuzzy subtractive clustering inversion for data 2.
International Journal of Geophysics 9
1.2
1.4
1
0
0.8
0.6
0.4
0.2
10 0.80.60.40.2
Field dataCluster centersSynthetic data
(a)
1.2
1.4
1
0
0.8
0.6
0.4
0.2
10 0.80.60.40.2
Field data
Cluster centersSynthetic data
(b)
1.2
1.4
1
0
0.8
0.6
0.4
0.2
10 0.80.60.40.2
Field data
Cluster centersSynthetic data
(c)
Figure 8: Formation of fuzzy cluster centers while interpreting data 2: (a) fuzzy πΆ-means clustering, (b) fuzzy πΎ-means clustering, and (c)fuzzy subtractive clustering.
accurate results because the adjustment of cluster centersand updating it. Therefore the overall result proves thatfuzzy subtractive clustering performance is satisfactory forthis problem under consideration. This may cause loss oforiginal information of the field data. Therefore the overallresult proves that fuzzy subtractive clustering performance issatisfactory.
7. Conclusion
Three clustering algorithms, namely, fuzzy πΆ-means clus-tering, fuzzy πΎ-means clustering, and fuzzy subtractiveclustering, have been applied in this paper for the inversionof geoelectrical resistivity data. These approaches solve theproblem in categorizing the results to predict the appropriate
10 International Journal of Geophysics
Figure 9: Output panel showing the fuzzy subtractive clustering inverted model for data 2.
93
93.5
94
94.5
95
95.5
96
96.5
0 50 100 150
Accu
racy
Number of iterations
y = 0.003x + 95.237
R2 = 0.0087
(a)
0 50 100 150Number of iterations
93
93.5
94
94.5
95
95.5
96
96.5
97
Accu
racy
R2 = 0.559
y = 0.0244x + 94.201
(b)
0 50 100 150Number of iterations
Best fitLinear (best fit)
95.5
96
96.5
97
97.5
98
98.5
Accu
racy
R2 = 0.8696
y = 0.0205x + 95.874
(c)
Figure 10: Performance measure of data 2 for the three algorithms: (a) fuzzyπΆ-means clustering, (b) fuzzyπΎ-means clustering, and (c) fuzzysubtractive clustering.
layer model. The FIS training provides enough syntheticdata necessary for framing the clusters. Three clusteringalgorithms have been implemented and tested with the fielddata based on the cluster centers and the results obtainedfrom all the three algorithms do not deviate from the earlierresults. It was obvious that from the inverted results thefuzzy πΆ-means algorithm fails to provide appropriate resultscompared to the other two algorithms. Subtractive clusteringseems to be a better performance algorithm on comparingwith fuzzy πΎ-means algorithm, since the density measure
calculated at each number of iterations is verymuchhelpful inmodelling the subsurface parameters. Finally, the clusteringtechniques discussed here in this paper can be used as astandalone approach for the interpretation of subsurface layermodel.
Conflict of Interests
The authors declare that there is no conflict of interestsregarding the publication of this paper.
International Journal of Geophysics 11
Acknowledgment
The authors would like to thank the reviewers for theirvaluable comments and suggestions, which have improvedthe paper.
References
[1] F. Aminzadeh and M. Jamshidi, Soft Computing: Fuzzy Logic,Neural Networks, and Distributed Artificial Intelligence, PrenticeHall PTR, Englewood Cliffs, NJ, USA, 1994.
[2] H. Flathe, βA practical method of calculating geoelectricalmodel graphs for horizontally stratified media,β GeophysicalProspecting, vol. 3, no. 3, pp. 268β294, 1955.
[3] W. K. Kosinski and W. E. Kelly, βGeoelectric soundings forpredicting aquifer properties,β Groundwater, vol. 19, no. 2, pp.163β171, 1981.
[4] M. Nikravesh and F. Aminzadeh, βMining and fusion ofpetroleum data with fuzzy logic and neural network agents,βJournal of Petroleum Science and Engineering, vol. 29, no. 3-4,pp. 221β238, 2001.
[5] U. Singh, D. Singh, and H. Singh, βApplication of Neurofuzzypattern recognition method in borehole geophysics,β Acta Geo-daetica et Geophysica Hungarica, vol. 45, no. 4, pp. 417β425,2010.
[6] Y. Srinivas, A. Stanley Raj, D. Hudson Oliver, D. Muthuraj, andN. Chandrasekar, βAn application of artificial neural networkfor the interpretation of three layer electrical resistivity datausing feed forward back propagation algorithm,βCurrent Devel-opment in Artificial Intelligence, vol. 2, no. 1, pp. 11β21, 2011.
[7] Y. Srinivas, A. Stanley Raj, D. Hudson Oliver, D. Muthuraj,and N. Chandrasekar, βEstimation of subsurface strata of earthusing Adaptive Neuro-Fuzzy inference system (ANFIS),β ActaGeodaetica et Geophysica Hungarica, vol. 47, no. 1, pp. 78β89,2012.
[8] Y. Srinivas, A. S. Raj, D. H. Oliver, D. Muthuraj, and N. Chan-drasekar, βA robust behavior of Feed Forward Back propagationalgorithm of artificial neural networks in the application of ver-tical electrical sounding data inversion,β Geoscience Frontiers,vol. 3, no. 5, pp. 729β736, 2012.
[9] S. Niwas and D. C. Singhal, βEstimation of aquifer transmissiv-ity from Dar-Zarrouk parameters in porous media,β Journal ofHydrology, vol. 50, no. C, pp. 393β399, 1981.
[10] M. Sugeno, Industrial Applications and Fuzzy Control, Elsevier,New York, NY, USA, 1985.
[11] P. M. Wong, F. Aminzadeh, and M. Nikravesh, Soft Computingfor Reservoir Characterization, Studies in Fuzziness and SoftComputing, Physica, Heidelberg, Germany, 2001.
[12] L. A. Zadeh, βFuzzy sets,β Information and Computation, vol. 8,pp. 338β353, 1965.
[13] L. A. Zadeh, βFuzzy logic, Neural networks, and soft comput-ing,βCommunications of the ACM, vol. 37, no. 3, pp. 77β84, 1994.
[14] U. Nordlund, βFormalizing geological knowledgeβwith anexample of modeling stratigraphy using fuzzy logic,β Journal ofSedimentary Research, vol. 66, no. 4, pp. 689β698, 1996.
[15] S. Cuddy, βThe application of mathematics of fuzzy logic topetrophysics,β in Proceedings of the SPWLA Annual LoggingSymposium, Paper S, 1997.
[16] J. H. Fang andH. C. Chen, βFuzzymodelling and the predictionof porosity and permeability from the compositional and
textural attributes of sandstone,β Journal of Petroleum Geology,vol. 20, no. 2, pp. 185β204, 1997.
[17] Y. Huang, T. D. Gedeon, and P.M.Wong, βPractical fuzzy inter-polator for prediction of reservoir permeability,β in Proceedingsof theIEEE International Fuzzy Systems Conference (FUZZ-IEEEβ99), Seoul, Republic of Korea, August 1999.
[18] P. Bois, βSome applications of pattern recognition to oil andgas exploration,β IEEE Transactions on Geoscience and RemoteSensing, vol. 21, no. 4, pp. 687β701, 1983.
[19] P. Bois, βFuzzy seismic interpretation,β IEEE Transactions onGeoscience and Remote Sensing, vol. 22, no. 6, pp. 692β697, 1984.
[20] J. C. Bezdek, Pattern Recognition with Fuzzy Objective FunctionAlgorithms, Plenum Press, New York, NY, USA, 1981.
[21] P. R. Krishnaiah and L. N. Kanal, Classification Pattern Recog-nition and Reduction of Dimensionality, vol. 2 of Handbook ofStatistics, North-Holland, Amsterdam, The Netherlands, 1982.
[22] J. Makhoul, S. Roucos, and H. Gish, βVector quantization inspeech coding,β Proceedings of the IEEE, vol. 73, no. 11, pp. 1551β1588, 1985.
[23] C. Chinrungrueng, Evaluation of heterogeneous architectures forartificial neural networks [Ph.D. thesis], University of California,Berkeley, Calif, USA, 1993.
[24] J.Moody andC. J. Darken, βFast learning in networks of locally-tuned processing units,β Neural Computation, vol. 1, no. 2, pp.281β294, 1989.
[25] S. L. Chiu, βFuzzy model identification based on cluster estima-tion,β Journal of Intelligent and Fuzzy Systems, vol. 2, pp. 267β278, 1994.
[26] K. S. Banerjee, S. P. Sharma, A. K. Sarangi, and D. Sengupta,βDelineation of subsurface structures using resistivity, VLFand radiometric measurement around a U-tailings pond andits hydrogeological implication,β Physics and Chemistry of theEarth, Parts A/B/C, vol. 36, no. 16, pp. 1345β1352, 2011.
[27] A. G. Batte, A. Muwang, and W. P. Sigrist, βEvaluating the useof vertical electrical sounding as a groundwater explorationtechnique to improve on the certainty of borehole yield inKamuli district (Eastern Uganda),β African Journal of Scienceand Technology, vol. 9, pp. 72β85, 2008.
Submit your manuscripts athttp://www.hindawi.com
Hindawi Publishing Corporationhttp://www.hindawi.com Volume 2014
ClimatologyJournal of
EcologyInternational Journal of
Hindawi Publishing Corporationhttp://www.hindawi.com Volume 2014
EarthquakesJournal of
Hindawi Publishing Corporationhttp://www.hindawi.com Volume 2014
Hindawi Publishing Corporationhttp://www.hindawi.com
Applied &EnvironmentalSoil Science
Volume 2014
Mining
Hindawi Publishing Corporationhttp://www.hindawi.com Volume 2014
Journal of
Hindawi Publishing Corporation http://www.hindawi.com Volume 2014
International Journal of
Geophysics
OceanographyInternational Journal of
Hindawi Publishing Corporationhttp://www.hindawi.com Volume 2014
Journal of Computational Environmental SciencesHindawi Publishing Corporationhttp://www.hindawi.com Volume 2014
Journal ofPetroleum Engineering
Hindawi Publishing Corporationhttp://www.hindawi.com Volume 2014
GeochemistryHindawi Publishing Corporationhttp://www.hindawi.com Volume 2014
Journal of
Atmospheric SciencesInternational Journal of
Hindawi Publishing Corporationhttp://www.hindawi.com Volume 2014
OceanographyHindawi Publishing Corporationhttp://www.hindawi.com Volume 2014
Advances in
Hindawi Publishing Corporationhttp://www.hindawi.com Volume 2014
MineralogyInternational Journal of
Hindawi Publishing Corporationhttp://www.hindawi.com Volume 2014
MeteorologyAdvances in
The Scientific World JournalHindawi Publishing Corporation http://www.hindawi.com Volume 2014
Paleontology JournalHindawi Publishing Corporationhttp://www.hindawi.com Volume 2014
ScientificaHindawi Publishing Corporationhttp://www.hindawi.com Volume 2014
Hindawi Publishing Corporationhttp://www.hindawi.com Volume 2014
Geological ResearchJournal of
Hindawi Publishing Corporationhttp://www.hindawi.com Volume 2014
Geology Advances in