Vous êtes sur la page 1sur 21

Multiple Criteria Decision Making using Generalized Data Envelopment Analysis

Y.B. Yun1, H. Nakayama2, M. Arakawa1, H. Ishikawa1


1 Faculty of Engineering, Kagawa University, Japan 1 2 Faculty of Science & Engineering, Konan University, Japan 2

E-mail:{yun,arakawa,ishikawa}@eng.kagawa-u.ac.jp, nakayama@konan-u.ac.jp E-mail:{yun,arakawa,ishikawa}@eng.kagawa-u.ac.jp, nakayama@konan-u.ac.jp


CSM'2001:15th JISR-IIASA Workshop, August 27-29, 2001

Purpose of This Research


to suggest the method for supporting decision makers in multiple criteria decision making (MCDM) to show an effectiveness of the proposed method through a numerical example
2001/08/28 Multiple Criteria Decision Making using GDEA 2

Contents
Background The Proposed method Numerical example Conclusion
2001/08/28 Multiple Criteria Decision Making using GDEA 3

Multiple Criteria Decision Making


min f ( x) = ( f1 ( x ),L ,f m ( x) )
x T

s.t. x X = { x R k|g j ( x) 0, j = 1,L ,l} x : design variable, f : criteria function

Pareto optimal solutions

Interactive optimization methods


ex) preferential optimization satisisficing quasi-satisficing

Final solution of decision making


2001/08/28 Multiple Criteria Decision Making using GDEA 4

MCDM by Interactive Optimization


solving an auxiliary optimization problem to obtain Pareto optimal solutions closest to the given aspiration level revising aspiration levels by making trade-off analysis
2001/08/28 Multiple Criteria Decision Making using GDEA 5

Multi-Objective Optimization Requiring Analysis


In engineering design problems, f cannot be given explicitly f(x) is obtained by some analyses such as structural, thermodynamic and fluid mechanical analysis These analyses take expensive computation time It takes too much time to obtain a Pareto optimal solution closest to the aspiration level Practically, interactive optimization methods are useless
2001/08/28 Multiple Criteria Decision Making using GDEA 6

Optimization by Genetic Algorithms


objective function = 2 or 3, GA is effective for figuring efficient frontiers
trade-off analysis based on visualized efficient

objective function 4, it is difficult to figure Pareto optimal solutions


decide a final decision making solution

To suggest an

aspiration level approach to GDEA


for supporting a decision making
2001/08/28 Multiple Criteria Decision Making using GDEA 7

Main Process of the Method


Step 1 Generation of Pareto Optimal
Initialization Crossover Mutation Fitness by GDEAAL Selection

Step 2

Step 3
2001/08/28 Multiple Criteria Decision Making using GDEA 8

Fitness Function
(GDEAAL)

2001/08/28

Multiple Criteria Decision Making using GDEA

Geometric Interpretation
f2
f (X )

y1

f : aspiration level
y2

f* : ideal point

f1
10

2001/08/28

Multiple Criteria Decision Making using GDEA

Simple Example

( min f1 ( x ), f 2 ( x ) ) = ( x1 , x 2 )
x

s.t. x1 3 x1 x 2 0
3

x1 1, x 2 2

Fig. 1. Population at the 30 generation

Fig. 2. Pareto optimal values obtained until 30 generation

2001/08/28

Multiple Criteria Decision Making using GDEA

11

Main Process of the Method


Step 1 Generation of Pareto Optimal
Initialization Crossover Mutation Fitness by GDEAAL Selection

Step 2 Selection of Alternatives


For x j being of the set Pareto otpimal solutions generated in Step 1, fi ( x j ) f i )0 )0 (1) A0 = f ( x ), where x = arg min max * x j i =1,L , p f i fi ) ) ( 2 ) Ai = f ( x i ), where x i = arg min f i ( x j ) f i , i = 1, L , p
xj

Step 3
2001/08/28 Multiple Criteria Decision Making using GDEA 12

Alternatives
f2
f (X )

A2 A0

f : aspiration level
A1

f* : ideal point

f1
13

2001/08/28

Multiple Criteria Decision Making using GDEA

Main Process of the Method


Step 1 Generation of Pareto Optimal
Initialization Crossover Mutation Fitness by GDEAAL Selection

Step 2 Selection of Alternatives


For x j being of the set Pareto otpimal solutions generated in Step 1, fi ( x j ) f i )0 )0 (1) A0 = f ( x ), where x = arg min max * x j i =1,L , p f i fi ) ) ( 2 ) Ai = f ( x i ), where x i = arg min f i ( x j ) f i , i = 1, L , p
xj

Step 3 Trade-off Analysis


2001/08/28 Multiple Criteria Decision Making using GDEA 14

Robust Design Problem


This problem is cited from 1) K.M. Ragsdell and D.T. Phillips, 1976. or 2) S.S. Rao, Engineering Optimization, 1996

h+hR

P (26689N) t

l+lR

L (=0.3556m) Fig. Cantilever Beam b

* Design Variables
hdepth of the weldllength of the weldtheight of the beam bthicknessof the beam h R error of the weld's depth l R error of the weld's length

2001/08/28

Multiple Criteria Decision Making using GDEA

15

Formulation
This is reformulated on the basis of Arakawa et als, 1998

Objective functions :
68216.1h 2 l + 2970.8tb( L + l ) min 136432.2hlh R + (68216.1h 2 + 2970.8tb)l R min 2970.8tbLR min h R max l R max
where, = + l + 2 , = R

Constraints :
: g1 shearing stress= a : gstress= a 2 : gdeviation= a 3 : gbuckling load= P Pc 4 g 5 := h R + h < b
P 2hl , = MR J

g 6 := h R h g 7 := l R l

l 2 (h + b) 2 l 2 + (h + b) 2 M = P(L + l/ 2 ), R = , J = 2hl + 6 4 4 3 3 6 PL 6 PL3 4.013 EI 1 t EI , I = tb , = Gtb = 2 , = 3 , Pc = 2L 12 3 t b Et b L2 P = 26689.3 (N), L = 0.3542 (m), E = 206.843 (GPa), G = 82.7371 (GPa), LR = 0.01 L (m)

2001/08/28

Multiple Criteria Decision Making using GDEA

16

Assumptions
Until 30 generation Ideal Point f * = (2.08836, 0.3688, 0.01381, 0.00205, 0.07535) Aspiration Level f = (2.2350, 0.450, 0.0150, 0.001, 0.006)

2001/08/28

Multiple Criteria Decision Making using GDEA

17

Alternatives to Aspiration Level


A0 0 -0.2 -0.4 -0.6 -0.8 -1
f f 1 f2 f 2 f3 f 3 f 4 f4 f 5 f5 * ( f1' , f2' , f3' , f4' , f5' , ) := 1 f f* , f f* , f f* , f* f , f* f 1 2 3 4 5 2 3 4 5 1

A1

A2

A3

A4

A5

f1' f2' f3' f4' f5'

Suppose improving f2 at A0
2001/08/28 Multiple Criteria Decision Making using GDEA 18

Trade-off Analysis
A0 A1 0
A2

f4 0.2 f5 0.25

-0.2 -0.4 -0.6 -0.8 -1

f1' f2' f3' f4' f5'

f f 1 f2 f 2 f3 f 3 f 4 f4 f 5 f5 * ( f1' , f2' , f3' , f4' , f5' , ) := 1 f f* , f f* , f f* , f* f , f* f 1 2 3 4 5 2 3 4 5 1

2001/08/28

Multiple Criteria Decision Making using GDEA

19

Improvement of Alternatives
A0
0 -0.2 f1' -0.4 -0.6 -0.8 -1
f f 1 f2 f 2 f3 f 3 f 4 f4 f 5 f5 * ( f1' , f2' , f3' , f4' , f5' , ) := 1 f f* , f f* , f f* , f* f , f* f 1 2 3 4 5 2 3 4 5 1

f2' f3' f4' f5'

2001/08/28

Multiple Criteria Decision Making using GDEA

20

Conclusions
Through this study, the following results are obtained: By the proposed method, it is possible to 1. 2. 3. helpful for supporting decision making make trade-off analysis applicable to wide optimization problems

2001/08/28

Multiple Criteria Decision Making using GDEA

21

Vous aimerez peut-être aussi