Proceedings of the Fifth International Conference on Machine Learning and Cybernetics, Dalian, 13-16 August 2006

AN IMPROVED EVOLUTIONARY PROGRAMMING ALGORITHM FOR

FUZZY PROGRAMMING PROBLEMS AND ITS APPLICATION
WEI-YI QIAN, JIN ZHANG

Department of Mathematics, Bohai University, Jinzhou, 121000, China

E-MAIL: qianweiyi@eyou.com

Abstract: exerted on the individual via other individuals. The fitness

An improved evolutionary programming algorithm is function is given by membership functions.
proposed for fuzzy programming problems. The idea is to An optimal control model of designing 3D trajectory
mimic the physics of electromagnetism by considering each in horizontal wells is formulated in a fuzzy environment.
individual as an electrical charge. Charge of each individual is Here, objective is to minimize the length of the trajectory
defined using membership functions. After calculating these
charges, like the electromagnetic force, we calculate a
and constraint is error of entry target point. The objective
combination force exerted on each individual. This and constraint are fuzzy in nature. The fuzzy objective and
combination force is regarded as the search direction of fuzzy constraint are defined by linear membership functions.
mutation operator. The fitness function is given by The model can be solved using the fuzzy nonlinear
membership functions. A fuzzy model is developed for programming and our proposed method. The model is
designing 3D trajectory of horizontal wells. The fuzzy model illustrated numerically and the results are efficiently.
can be transformed to fuzzy programming problem and solved
by our proposed method. The model is illustrated numerically 2. Fuzzy Programming
and the results are efficiently.
We consider fuzzy programming problem in the
Keywords: following form:
Fuzzy programming; membership function; horizontal
~
wells; evolutionary programming algorithm
min f ( x)
~
1. Introduction s.t. g i ( x) ≤ bi i = 1,2, L , m
(1)
l ≤ x≤u
Fuzzy mathematical programming has been applied to
where x, l and u ∈ R n , are an n-dimensional vector, f
several fields like project network [1], transportation
problem [2], air pollution regulation [3], inventory problem and g j ( j = 1,2, L m ), are a real function.
[4], etc. The fuzzy mathematical programming can be Let a is the target to be achieved by the objective. In
transformed to a regular crisp nonlinear programming the fuzzy theory, the fuzzy objective and the fuzzy
problem using fuzzy set theory. There are some inherent constraints are defined by linear membership functions.
difficulties to solve the crisp problems by conventional
Here, assumed µ f and µ g j to be the linear membership
mathematical tools, such as gradient methods. To overcome
these difficulties, recently stochastic search methods are functions for fuzzy objective and constraints, respectively.
used as optimization techniques for the crisp problems such These are represented as
as genetic algorithms, etc. A stochastic search algorithm 1 if f ( x) ≤ a
may provide a useful solution when other algorithms fail. 
A new improved evolutionary programming algorithm  f ( x) − a
µ f ( f ( x)) = 1 − if a ≤ f ( x) ≤ a + p
is developed for nonlinear programming problems with  p
fuzzy objective and fuzzy constraints. In this method, 0 if f ( x) > a + p
charge of each individual is defined using membership
function. After calculating these charges, based on idea of
[5], we use them to find a direction of mutation operator.
We select the direction by evaluating a combination force

1-4244-0060-0/06/$20.00 ©2006 IEEE

1851

Authorized licensed use limited to: UNIVERSITI TEKNOLOGI MARA. Downloaded on July 15,2023 at 05:22:34 UTC from IEEE Xplore. Restrictions apply.
 Proceedings of the Fifth International Conference on Machine Learning and Cybernetics, Dalian, 13-16 August 2006

1 if g j ( x) ≤ b j Hooke-Jeeves method [7]. The x i is replaced by the optimal


 g j ( x) − b j solution y i .
µ g j ( g j ( x)) = 1 − if b j ≤ g j ( x) ≤ b j + d j
dj Mutation operation: Mutation operator is main operator

0 in improved evolutionary programming algorithm. For each
 if g j ( x) > b j + d j
mutated individual x i , its charge is defined as follows.
j = 1,2, L , m
q i = exp( βµ min ( x i )), i = 1,2, L , N
where p and d j ( j = 1,2, L , m) are the maximally
where β ≥ 1 is a given constant
acceptable violations of the aspiration levels of a and b j . In this way, individuals that have higher membership
Using the fuzzy nonlinear programming technique, the degree possess higher charges.
problem (1) can be formulated as Notice that, unlike electrical charge, no signs are
max λ attached to the charge of a sample point. Instead, we decide
the direction of a particular force between two individuals
f ( x) − a after comparing their membership degree. Hence, the total
s.t. 1 − ≥λ
p force exerted on individual x i is computed by the
(2)
g j ( x) − b j following
1− ≥ λ j = 1,2, L , m
dj  j i qiq j
( x − x ) 2
if µ min ( x j ) > µ min ( x i )
0 ≤ λ ≤ 1, l ≤ x ≤ u N  x j − xi
 (4)
Now, it is a crisp deterministic problem. We solve this F ( x (t )) = ∑ 
i

j ≠i  j i qiq j
problem by improved evolutionary programming algorithm ( x − x ) 2
if µ min ( x j ) ≤ µ min ( x i )
given in section 3.  x j − xi

3. Improved Evolutionary Programming Algorithm Let
F ( x i (t ))
Fi = ,
The improved evolutionary programming algorithm F ( x i (t ))
starts with an initial population of N individuals: an
individual is composed of real coordinates. The main the new component x ki of the individual x i is performed as
procedure of the algorithm can be described as follows: follows:
1. Generation of initial population;
2. Local search; i i
λFki (u k − x ki ) if Fki > 0
xk = xk + 
3. Mutation operation; λFki ( x ki − l k ) else
4. Selection operation;
5. If stopping criteria is satisfied, then stop; else go to 2 where λ is a uniformly distributed random variable, Fki
We stress the basic elements of the algorithm as u k and l k are the kth component of the corresponding
follows.
Generation of initial population: To cover the whole vector, respectively.
solution space, and to avoid the risk of having too much Selection operation: The selection method is elitist
individuals in the same region, the algorithm selects selection. In other words, the best N individuals are
Uniform Design method [6]. reserved.
Stopping criteria: In our study, we terminate the
Fitness function: For each individual x i , the fitness algorithm procedure by using a maximum number of
function is defined as follows iterations.
Fit ( x i ) = µ min ( x i ) (3)
where 4. Fuzzy Model
µ min ( x k ) = min{µ f ( f ( x k )), µ g j ( g j ( x k )), j = 1,2, L , m}
4.1. Crisp Model
Local search: For each individual x i , the problem
A crisp optimal control model of designing 3D
max µ min ( x) is solved in neighborhood of x i by using
trajectory of horizontal wells is given in [8] as follows:

1852

Authorized licensed use limited to: UNIVERSITI TEKNOLOGI MARA. Downloaded on July 15,2023 at 05:22:34 UTC from IEEE Xplore. Restrictions apply.
 Proceedings of the Fifth International Conference on Machine Learning and Cybernetics, Dalian, 13-16 August 2006

min J = sn n
min J = ∑ u i 3
dxi1 cos ui 2
s.t. = i =1
ds ui1 5
dxi 2
=
sin ui 2 s.t. ∑ x nj ( z , x 0 ) − x tj < ε (6)
j =1
ds ui1 sin xi1 ( s )
dxi 3 a ij ≤ u ij ≤ bij
= sin xi1 ( s ) cos xi 2 ( s ) s ∈ ( si −1 , si ],
ds
dxi 4 4.2. Fuzzy Model
= sin xi1 ( s ) sin xi 2 ( s ) i = 1,2,L, n (5)
ds
When the above objective and constraints become
dxi 5
= cos xi1 ( s ) fuzzy, the said crisp model is transformed to
ds
xij ( si −1 ) = xi −1 j ( si −1 ) i = 1,2,L, n j = 1,2,L,5 ~ n
min J = ∑ u i 3
x1 j (0) = x0 j i =1
5 5
~ε
∑x nj ( sn ) − xtj < ε s.t. ∑ x nj ( z, x 0 ) − xtj <
(7)
j =1 j =1

aik ≤ uik ≤ bik i = 1,2,L, n k = 1,2 a ij ≤ u ij ≤ bij

where the unit of length and angel are meter and radian, The problem (7), like section 2, can be formulated as
respectively. The independent variable s is the arc-length of max λ
the trajectory, si is the arc-length from the start point of n

the trajectory to endpoint of the i th curve segment, s 0 ∑ u i3 − L

i =1
(=0) is the arc-length at start point of the trajectory. s.t. 1 − ≥λ
p
xi1 , xi 2 , xi 3 , xi 4 and xi 5 denote the inclination, azimuth, 5
north coordinate, east coordinate and vertical coordinate on ∑ x nj ( z , x 0 ) − x tj − ε
the i th curve segment, respectively. u i1 and u i 2 , the 1−
j =1
≥λ
curvature radius and the tool-face angle on the i th curve d (8)
segment, are the constant x t1 , xt 2 , xt 3 , x t 4 and x t 5 are 0 ≤ λ ≤ 1, a ij ≤ u ij ≤ bij
the inclination, azimuth, north coordinate, east coordinate The problem (8) is solved using improved
and vertical coordinate at target point. ε (> 0) is a evolutionary programming algorithm given in section 3.
tolerance. a ij and bij are the lower bound and the upper 5. Numerical Example
bound of the variable u ij . n denotes a number of curve
To illustrate model, we take the well Zi16-146 (a short
segments
radius well) for example. Assume n = 3 , x 01 = 10.4 o ,
Define u i 3 = s i − s i −1 , x ij = xij ( s i ), x 0j = x1 j (0) ,
x 02 = 228.18 o , x 03 = 102.69m , x 04 = −156.39m , x05 =
x 0 = ( x10 , L , x 50 ) T ∈ R 5 , and x i = ( x1i , x 2i , L , x 5i ) T ∈ R 5
Choosing the optimization variable 1678.15m , x t1 = 89.5 o , x t 2 = 205.5 o , x t 3 = 62.5m ,
z = (u11 , u12 , u13 , L , u n1 , u n 2 , u n3 ) T ∈ R 3n , x t 4 = −192.9m , x t 5 = 1718.0m , a = 60m , ε = 0.5m ,
we integrate the differential equations above and then a11 = a 21 = a 31 = 40m , b11 = b21 = 31 = 60m , a12 = a22 =
i
compute recursively the variables x , a 32 = −50 o , b12 = b22 = 32 = −50 o , a13 = a23 = a 33 = 10m ,
x i = x i ( z , x 0 ) = x i (u11 , u12 , u13 , L , u i1 , u i 2 , u i 3 , x 0 ) b13 = b23 = 33 = 100m . The optimal results of crisp model
are functions of the optimization variable z . Hence the for the Zi16-146 well are presented in table 1 using general
optimal control model (5) can be formulated as follows: evolutionary programming algorithm. TFA, CR, AL and TL
in the table 1 denote the Tool-face angle, curvature radius,

1853

Authorized licensed use limited to: UNIVERSITI TEKNOLOGI MARA. Downloaded on July 15,2023 at 05:22:34 UTC from IEEE Xplore. Restrictions apply.
 Proceedings of the Fifth International Conference on Machine Learning and Cybernetics, Dalian, 13-16 August 2006

arc length of each segment and total length of trajectory, Acknowledgements

respectively. If L = 70m , p = 40m , d = 0.3m , β = 2 ,the
optimal results of fuzzy model for the Zi16-146 well are This paper is supported by the Educational Department
presented in table 2 using the improved evolutionary Foundation of Liaoning province (NO:200540).
programming algorithm. The maximum number of itera-
tions is 4000 in two algorithms. References

Table 1. Optimal results of the crisp model [1] Deporter E.L. and Ellis K.P., Optimization of project
First seg. Second seg. Third seg. TL network with goal programming and fuzzy linear
programming, Computers and Industrial Engineering,
TFA 15.780081 -25.510206 -49.000213 1990,19:500-504.
CR 50.306272 49.480754 45.816337 [2] Bit A.K., Biswal M.P. and Alam S.S., Fuzzy
AL 17.873991 34.387768 24.693885 76.955644
programming approach to multi criteria decision
making transportation problem, Fuzzy set and Systems,
1992,50: 135-141.
Table 2. Optimal results of the fuzzy model [3] Sommer G. and Pollatschek M.A., A fuzzy
First seg. Second seg. Third seg. TL programming approach to an air pollution regulation
TFA 4.369715 -11.930033 0.303889 problem, European Journal of Operational Research,
1978,10:303-313.
CR 47.184242 55.956892 54.813648 [4] Roy T.K. and Maiti M., Mulit-objective inventory
AL 28.508348 11.495308 35.158369 75.162025 models of deteriorating items with some constraints in
a fuzzy environment, Computers and Operations
Research, 1998, 25:1085-1095.
The numerical results illustrate the accuracy and [5] Ilker Birbil S. and Fang S.C., A multi-point stochastic
efficiency of the fuzzy model and the improved search method for global optimization, Operation
evolutionary programming algorithm Research and Its Applications, World Publishing
Corporation, Beijin, 2002.
6. Conclusions [6] Fang K. and Ma C. Orthogonal and Uniform
Experimental Design, Science Press, Beijing, 2001
In this paper, an improved evolutionary programming [7] Hooke R. and Jrrvers T.A., Dirrct search solution of
algorithm is developed for nonlinear programming numerical and statistical problems Computer Math.,
problems with fuzzy objective and fuzzy constraints. An 1961, 8:212-229.
optimal control model of designing 3D trajectory in [8] Qian W.Y., Improved evolutionary programming
horizontal wells is developed in a fuzzy environment. The method of nonlinear optimal control model for
fuzzy model is applied to practical design of the 3D designing three dimensional trajectory of horizontal
horizontal well and solved by our proposed improved wells, Journal of the University of Petroleum Chain,
evolutionary programming algorithm. 2003,27(6): 30-32.

1854

Authorized licensed use limited to: UNIVERSITI TEKNOLOGI MARA. Downloaded on July 15,2023 at 05:22:34 UTC from IEEE Xplore. Restrictions apply.