You are on page 1of 5

JOURNAL OF COMPUTING, VOLUME 2, ISSUE 9, SEPTEMBER 2010, ISSN 2151-9617

HTTPS://SITES.GOOGLE.COM/SITE/JOURNALOFCOMPUTING/
WWW.JOURNALOFCOMPUTING.ORG 14

Two Approaches for Fuzzy Interpolation


I.N. Askerzade Askerbeyli, Şahin Emrah Amrahov

Abstract—In this work a generalized method of linear and cubic spline interpolation of fuzzy data is presented. Least square
method (LSM) and geometric approach to the fuzzy numbers are used. Examples of application linear spline interpolation for
parametric fuzzy numbers and cubic spline for triangular fuzzy numbers are presented.

Index Terms—Interpolation formulas, Least squares methods, Linear approximation, Spline and piecewise polynomial
interpolation

——————————  ——————————

1 INTRODUCTION

I T is important to generate new data from the data that


we have. Although initial information is certain the
data we are going to generate will often be approx-
2 A LINEAR SPLINE MODEL FOR FUZZY
DATA POINTS

imate. When there exists uncertinaty in data, our results Firstly we will consider points with crisp coordinates. In
are not only approximate, but also uncertain. As is known linear spline S ( x) passing through the data points
cases involving uncertainty can be modeled in three dif-
1
ferent way mathematically. For this, probability theory, ( x1 , y1 ) and ( x2 , y2 ) given by the expression,
interval analysis or fuzzy logic can be used. Fuzzy logic
is developed bu Zadeh at 1965 and the theory is still im- S1 (x)  a1 x  b1 for x  [ x1 , x2 ) (1)
proving. Fuzzy interpolation problem is a part of this
theory and it is one of the problems of this theory that has Similarly, when x  [ x2 , x3 ) , the linear spline S2 ( x) pass-
to be improved. Zadeh mentioned interpolation problem
ing through ( x2 , y2 ) and ( x3 , y3 ) is used
of fuzzy logic in his first article [16]. The first studies
about this topic are belongs to Lowen [12] and Kaleva S2 (x)  a2 x  b2 (2)
[10]. System of equations
Splines are used in many areas in classical mathematic. y1  S1 (x1 )  a1x1  b1
In fuzzy mathematic fuzzy spline approach from crisp y2  S1 (x2 )  a1 x2  b1
data is shown by Kaleva. Abbasbandy and Babolian has
learned natural and complete spline topic for fuzzy data
y2  S2 (x2 )  a2 x2  b2 (3)
[1-2]. B-splines are applied to a model by Anile et.al. [3] y3  S2 (x3 )  a2 x3  b2
and theory of them are developed by Blaga and Bede [4]. gives expression for coefficients a1 , a2 , b1 , b2 . Similar result
In all these papers inputs are crisp and function values can be obtained by using Least Square Method [11,15]. In
are fuzzy. Gáti and Bede examined how to obtain B-
LSM method in the case fitting a straight line requires the
splines from fuzzy data by using Zadeh extension pren-
minimization of
siple [5-6].
In this paper we suggest two new approaches for fuzzy
 ( a, b )   ( y  ax  b)
m 2
splines. First of these approach is applied to first degree i 1
i i . (4)
splines. Second one is a geometric approach. Firstly crisp A necessary condition yield to the equations
centers are combined with natural splines and then fuzzy
  0,   0 (5)
degree is found by the help of geometric smilarity. In sec-
a b
tion 2 generalized Least Square Method was introduced
The solution for the parameters a and b are easily ob-
for linear spline of fuzzy points. In section 3 we discuss
tained by elimination and are found to be
possibility of cubic spline in the case fuzzy data points.
Examples of finding of coefficients of linear and cubic
spline interpolation are presented.
m  xi yi  x i yi
a  , (6a)
m  x i2  ( x ) i
2

————————————————
 I.N.Askerbeyli and S.E. Amrahov are with the Ankara Universi-
ty,Engineering Faculty,Computer Engineering Depart-
ment,Tangogan,Ankara,Turkey,06100
JOURNAL OF COMPUTING, VOLUME 2, ISSUE 9, SEPTEMBER 2010, ISSN 2151-9617
HTTPS://SITES.GOOGLE.COM/SITE/JOURNALOFCOMPUTING/
WWW.JOURNALOFCOMPUTING.ORG 15

x  y x y x 2
i i i i i
m 1
L   ( d(Ly,i ()Lx,i ()  Ry,i ()Rx,i ()) 
b (6b)
m x  ( x ) 2
i i
2 i 1 0
(15)
m 1
The case of m  2 correspond to the results which coin-   d(Lx,i ()  Rx,i ()))
cide with results of solution of system of equations (3). i 1 0
In the case of points with fuzzy coordinates, we will m 1
use triangular parametric number [11, 15], i.e. M  2m ( d(L2x,i ()  Rx2,i ()) (16)
i 1 0

L( )  1 ( ) and R ( )  1 ( ) ,


1
(7) m
N   ( d(L ()  Rx,i ()) 
2
x,i (17)
i 1 0

where 1 ( ) and 1 ( ) are inverse functions for in- Application of expressions (11)-(17) for each pairs of adja-
creasing left hand and deceasing right-hand sides of the cent fuzzy points leads spline interpolation. For each in-
membership function  ( x) , respectively. If use ap- terval we find the values of slopes and intersepts ai , bi .
proaches similar to (3), then we can get system of equa-
tions: Example: Consider the following triangular fuzzy points:

Ly1 ( )  a1Lx1 ()  b1 Lx1 ( )   ; Rx1 ( )  2   ; 0    1


Ry1 ( )  a1Rx1 ( )  b1 (8) Ly1 ( )  1   2 ; Ry1 ( )  3   2 ; 0    1
Ly2 ( )  a1Lx2 ( )  b1 Lx 2 ( )  2   2 ; Rx 2 ( )  4   ; 0    1
Ry2 ()  a1Rx2 ( )  b1 Ly 2 ( )  3   ; Rx 2 ( )  5   ; 0    1
Solution of last system of equations leads to contradic- Lx 3 ( )  4   2 ; Rx 3 ( )  6   2 ; 0    1
tion with crisp character of coefficients a1, b1 . For this
Ly 3 ( )  6   2 ; Rx 3 ( )  8   ; 0    1
reason we will use fuzzy least square method for linear
spline. For the generalization of LSM we use following Implementation of the general procedure given by
distance between fuzzy numbers A and B criterion pro- Esq.(11)-(17) yields the solution a1  0.823, b1  1.372 and
posed by Ma et al [13,14] a2  1.526, b2  0.457 .
1 1
D(A, B)    d(RA ()  RB ())2   d(LA ()  LB ())2  3 A CUBIC SPLINE MODEL FOR FUZZY DATA
1/2
(9)
0 0 POINTS
In similar way for fuzzy LSM we can write expression
Two fuzzy triangular numbers identify a rectangle in xy
m 1 coordinate space. For fuzzy triangular numbers
   (  d  (R
i 1
y ,i ( )  aR x , i ( )  b ) 2 
 x  a1 , b1 , c1  and  y  a 2 , b2 , c 2 
0
(10)
1 this rectangle is given by:
  d  (L y , i ( )  aL x , i ( )  b ))1/ 2
0
S  ( x, y ) a1  x  c1 , a 2  y  c 2 
Calculations based on expression (5) leads to results: In fact, this rectangle is a fuzzy set obtained from the Car-
tesian product of two fuzzy sets. The point
F E K L B  b1 , b2 
a , b (11)
M N M N has grade 1. Let us call the point B the crisp center of S.
We note that this relation between the pairs
where ( x ,  y )
m 1 m 1 and the pairs
E   ( d(Ly,i ()  Ry,i ()) ( d(Lx,i ()  Rx,i ()) (12) ( S , B) , where B a point on or inside the rectangle is S ,
i 1 0 i 1 0
is a bijection. Hence, one could interpret S as a fuzzy
1
rectangle with fuzzy coordinates
m
F  m ( d(Ly,i ()Lx,i ()  Ry,i ()Rx,i ()) (13) ( x ,  y ) and with crisp center B.
i 1 0
m 1 m 1
K   ( d(Ly,i ()  Ry,i ())  d(L2x,i ()  Rx2,i ())) (14)
i 1 0 i 1 0
JOURNAL OF COMPUTING, VOLUME 2, ISSUE 9, SEPTEMBER 2010, ISSN 2151-9617
HTTPS://SITES.GOOGLE.COM/SITE/JOURNALOFCOMPUTING/
WWW.JOURNALOFCOMPUTING.ORG 16

Step 3: We determine the fuzzy y coordinate of S as


follows:
We compute the distances between the crisp centers of Si
and Si 1 , and their lower and upper sides. We denote
these distances by:
yid , yiu , y(i 1) d , y(i 1)u respectively.
Let the distances from D to the lower and upper sides of
S be yd , yu respectively.
Then we order the numbers biy , f (b) and b(i 1) y . If
b(i 1) y  biy then
yd  yid y(i 1) d and
yu  yiu y(i 1)u
If b(i 1) y  biy and f (b)  biy then yd  yid , yu  yiu .
If b(i 1) y  biy and f (b)  b(i 1) y then
yd  y(i 1) d , yu  y(i 1)u .
This geometric interpretation of pairs of fuzzy numbers If b(i 1) y  biy  f (b) then let
was first suggested by N.Gasilov et. al. [7-8].
Now let fuzzy rectangles S1 , S 2 ,...S n given with crisp cen- M  max(b(i 1) y , biy ) and m  min(b(i 1) y , biy ) .
ters at Let
A1 (b1x , b1 y ) , A2 (b2 x , b2 y ) ,… An (bnx , bny ) ,  y(i 1) d , if M  b(i 1) y
yMd  
respectively. One could construct the cubic spline corres-  yid , if M  biy
ponding to S1 , S 2 ,...S n as follows: We define ymd , yMu and ymu in a similar fashion, namely
Step 1: We find the cubic spline passing through  y(i 1)u , if M  b(i 1) y
yMu  
A1 (b1x , b1 y ) , A2 (b2 x , b2 y ) ,… An (bnx , bny ) ,  yiu , if M  biy
Suppose this curve is given by y  f ( x) . The curve  y(i 1) d , if m  b(i 1) y
represents fuzzy rectangles with crisp centers on the ymd  
 yid , if m  biy
curve.
Step 2: The fuzzy triangular number on the x axis corres-  y(i 1)u , if m  b(i 1) y
ymu  
ponding to the fuzzy rectangle S with center at  yiu , if m  biy
D(b, f (b)) is determined as follows: If m  f (b)  M then
Let bix  b  b(i 1) x ( f (b)  m) yMd  ( M  f (b)) ymd
yd  (20)
We compute the distances between the crisp centers of M m
Si and Si 1 and their left and right sides. Let us denote and
these distances by: ( f (b)  m) yMu  ( M  f (b)) ymu
yu  (21)
xil , xir , x(i 1)l , x(i 1) r , respectively. M m
Let the distances between D and the left and right sides
of S be xl , xr , respectively. Then we have, If m  M  f (b) then
b(i 1) x  b x(i 1) r  xr
  f (b)  M
b  bix xr  xir  ymu  M m
( yMu  ymu ), if ymu  yMu
Hence, yu  
y  f (b)  M
(b  bix ) x(i 1) r  (b( i 1) x  b) xir ( ymu  yMu ), if ymu  yMu
xr  (18)  Mu M m
b( i 1) x  bix and
Similarly,  f (b)  M
(b  bix ) x(i 1)l  (b(i 1) x  b) xil  ymd  M  m ( yMd  ymd ), if ymd  yMd
xl  (19) yd  
b(i 1) x  bix  y  f (b)  M ( y  y ), if ymd  yMd
 Md M m
md Md

Therefore, the fuzzy x coordinate of S is the fuzzy tri-


angular number (b  xl , b, b  xr ) .
JOURNAL OF COMPUTING, VOLUME 2, ISSUE 9, SEPTEMBER 2010, ISSN 2151-9617
HTTPS://SITES.GOOGLE.COM/SITE/JOURNALOFCOMPUTING/
WWW.JOURNALOFCOMPUTING.ORG 17

1
x1l 
2
x1r  2
If f (b)  m  M then
x2l  1
x2 r  1
 m  f (b)
 ymu  M  m ( yMu  ymu ), if ymu  yMu From the equations (18)-(19)
yu   x2 r  x1r 3
 y  m  f (b) ( y  y ), if ymu  yMu xr  
 M m 2 2
Mu mu Mu

and x2l  x1l 3


xl  
 m  f (b) 2 4
 ymd  M  m ( yMd  ymd ), if ymd  yMd Hence x-coordinate of the fuzzy point with with crisp
yd   center abscissa 2 is triangular number
 y  m  f (b) ( y  y ), if ymd  yMd 5 7
 Md M m
md Md
 , 2,  .
4 2
Therefore, the fuzzy y coordinate of S is the fuzzy tri-
Now we can calculate y- coordinate of this point. We
angular number have
( f (b)  y d , f (b), f (b)  y u ) y1d  1
y1u  2
Example.
y2 d  2
a) Find the cubic spline corresponding to following fuzzy y2 u  2
points Since 2  f (2)  16 from the (20) and (21)
1 9
K(( ,1,3),(1,2,4)), L((2,3,4),(14,16,18)) and yd 
2 7
9
M(( ,5,6),(60,62,63)) yu  2
4
b) Find the coordinates of the fuzzy point with crisp cen- Hence y-coordinate of the fuzzy point with crisp center
ter abscissa 2 on cubic spline ordinate 6 is triangular number
Solution.
 33 
a) A1 (1, 2), A2 (3,16) and A3 (5, 62)  ,6,8  .
are the crisp centers of the given fuzzy points.  7 
Let f ( x ) be natural cubic spline crossing through these
points. We have
4 CONCLUSIONS
a x 3  a2 x 2  a1 x  a0 , if x  1,3
f ( x)   3 3
 b3 x  b2 x  b1 x  b0 , if x  3,5
2
In this paper we discuss how we can apply a general
We have the following conditions: LSM for linear spline and geometric approach for cubic
a3  a2  a1  a0  2 spline of fuzzy data points. Model treats all fuzzy num-
bers which are presented by representation L-R parame-
27 a3  9a2  3a1  a0  16 tric way. In fuzzy number space, a metric is defined using
27 a3  6a2  a1  27b3  6b2  b1 Ma approximation. Examples of fuzzy linear spline and
18a3  2a2  18b3  2b2 cubic spline are presented.
27b3  9b2  3b1  b0  16 References
125b3  25b2  5b1  b0  62
[1] S. Abbasbandy, Interpolation of fuzzy data by
6a3  2a2  0 complete splines, Korean J. Comput. & Appl. Math., Vol
8, No 3, pp. 587-594,2001
30b3  2b2  0
[2] S. Abbasbandy, E. Babolian, Interpolation of fuzzy data
Solving these equations we find by natural splines, Korean J. Comput.& Appl. Math., Vol. 5,
 x3  3 x 2  6 x  2, if x  1,3 pp. 457-463, 1998
f ( x)   3 [3] A.M. Anile, B. Falcidieno, G. Gallo, M. Spagnuolo,S.
 x  15 x  48 x  52, if x  3,5
2
Spinello, Modeling uncertain data with fuzzy B-splines,
b) We have f (2)  6 Fuzzy Sets and Systems, Vol 113, pp. 397-410,2000
JOURNAL OF COMPUTING, VOLUME 2, ISSUE 9, SEPTEMBER 2010, ISSN 2151-9617
HTTPS://SITES.GOOGLE.COM/SITE/JOURNALOFCOMPUTING/
WWW.JOURNALOFCOMPUTING.ORG 18
[4] P. Blaga, B. Bede, Approximation by fuzzy B-spline
series, Journal of Applied Mathematics and Computing,
Vol 20 ,pp. 157-169,2006

[5] J. Gáti, B. Bede, Variation diminishing spline


approximation of fuzzy functions, WSEAS Transactions
on Mathematics, Vol 5, pp. 48-53,2006
[6] J. Gáti, B. Bede, Spline approximation of fuzzy
functions, Proceedings of the 8th WSEAS International
Conference on Applied Mathematics, Tenerife, Spain,
pp.194-199, December 16-18,2005
[7] N.A. Gasilov, Sh.G. Amrahov, A.G. Fatullayev, H.I.
Karakasş, Ö. Akın, Existence theorem for fuzzy number
solutions of fuzzy linear systems. Proceedings of first In-
ternational Fuzzy Systems Symposium (FUZZYSS’09).
Ankara, Turkey. Edited by T. Dereli, A. Baykasoglu and
I.B. Turksen,pp. 436-439, October 1-2, 2009
[8] N.A. Gasilov, Sh.G. Amrahov, A.G. Fatullayev, H.I.
Karakasş, Ö. Akın, A geometric approach to solve fuzzy
linear systems. htpp://arxiv.org/ftp/arxiv/ pa-
pers/0910/0910.4049.pdf
[9] R.Giordano, M.D.Weir, W. P.Fox, A first course of ma-
thematical modeling, Thompson Publisher, 2003
[10] O. Kaleva, Interpolation of fuzzy data, Fuzzy Sets and
Systems, Vol. 61, pp. 63-70, 1994
[11] Kwang H.Lee., First course on Fuzzy Theory and
Applications, Springer, 2005
[12] R. Lowen, A fuzzy Lagrange interpolation theorem,
Fuzzy Sets and Systems Vol.34,pp. 33-38,1990
[13] M.Ma, A.Kandel, M.Friedman, A new approach for
defuzzfication, Fuzzy Sets and Systems, Vol. 111, pp.351-
356,2000
[14] M.Ma, A.Kandel, M.Friedman, Correction to ‘’A new
approach for defuzzfication’’, Fuzzy Sets and Systems,
Vol. 128,pp. 133-134,2002
[15] E.Nasibov, S.Peker, On the parametric approximation
of a fuzzy number, Fuzzy Sets and Systems, Vol. 159, pp.
1365-1375, 2008
[16]L.A. Zadeh , Fuzzy sets, Inform. And Control, Vol. 8 ,
pp.338-353, 1965

You might also like