Você está na página 1de 6

LECTURE 3

OUTRANKING METHODS-- Outranking methods


PROMETHEE Enrichment of dominance relation
adding arcs to dominance relation
A. Main Approaches building fuzzy dominance relations
Aggregation of criteria Use of outranking relation as
Multicriteria value replaced by single decision aid
criterion
Complete dominance relation
Interactive method
Decision maker preferences not modeled
globally, but incrementally
1 2

B. PROMETHEE METHOD Evaluation Table


Preference Ranking Organization f1 () f 2 () f k ()
METHod for Enrichment a1 f1 (a1 ) f 2 ( a1 ) f k (a1 )
Evaluation (Brans et al., 1986) a2 f1 (a2 ) f 2 ( a2 ) f k ( a2 )
Problem
max { f1 (x ),..., f k ( x ) | x A}

Assume set A is finite set of an f1 (an ) f 2 ( an ) f k ( an )
potential alternatives

3 4

II f1 () f 2 ()
Examples
2 Criteria f1 () f 2 ()
a
b
100
30
20
100 } Incomparable
2 Actions a and b
Both a and b efficient
I f1 () f 2 () Decision maker must choose
a 100 100 III f1 () f 2 ()
b 30 20
a 100 99
b 20 100
Only a is efficientdominating f(b)
Both a and b efficient, but not incomparable
a preferred to b
5 6

1
IV f1 () f 2 () Summary
a
b
100
99
99
100 } Indifferent Efficiency True
Both a and b efficient, but negligible Criteria Conclusion
comparabilitydecision maker must choose I a effic. aPb
V f1 () f 2 () II a,b effic aRb
a 100 100 III a,b effic aPb
b 99 99 IV a,b effic aIb
V a effic aIb
Only a efficient, but really indifferent
b could be selectednot excluded from
decision making process
7 8

Requirements
1. Deviations between values of criteria must
C. PREFERENCE FUNCTIONS
be considered Measure of intensity of preference of a
2. Scaling effects eliminated over b
0 P ( a , b) 1
3. Pairwise comparisons should result in
aPb aIb aRb a, b A
i.e., partial preorder P ( a , b) = 0
4. Simple and understandable to decision No preference
maker d = f ( a ) f ( b) 0
5. No technical parameters P ( a , b) 0
Weak preference
6. Clearly display conflicting nature of d = f ( a ) f ( b) > 0
criteria
9 10

Example preference relation


P ( a , b) < 1 Strong
P(a, b) degree of truth that
d = f (a ) f (b) >> 0 preference
a preferred over b
1
P ( a , b) = 1 Strict
preference
d = f ( a ) f (b) >>> 0 fuzzy set

d = f (a) f (b)

11 12

2
Compound preference relations: Type I H (d )
P ( a , b) d 0 [Insensitive]
H (d ) = 1
P(b, a ) d 0
H (d )
1

Generalized d = f (a) f (b)


0 if d = 0
criterion H (d ) =
d = f ( a ) f ( b) 1 if d > 0
{ f (), P(a, b)} Corresponds to classical (I,P) situation
{ f (), H (d )} Insensitive to d
13 14

Type II H (d ) Type III H (d )


[Indifference] [Linear]
1 1
indifference
region
q +q d d
p +p
0 if d q d
H (d ) = if d p
H (d ) = p
1 if d > q 1 if d > p
User defines indifference threshold q
Linear increase in intensity with d
15 16

H(d )
Type IV H (d ) Type V
[Level 1 [Linear 1
Criterion] with
Indifference]
+ p q +q + p d
+p q +q +p d
0 if d q
0 if d q
H ( d ) = 12 if q < d p d q
1 if d > p H (d ) = if q < d p

2 parameters: p and q pq
1 if d > p
Good for finite order sets (e.g., bad, ave., Practical technique
good)
17 18

3
Type VI H (d ) D. FUZZY OUTRANKING
[Gaussian]
1
Assume for each f j () we have H j ()
or generalized criterion
{f j (), H j ()}
+
d 2 let k
H (d ) = 1 e 2 2 (a, b) = w j Pj (a, b)
Similar to Type V, but only 1 parameter k
j =1

wj = 1
j =1
19 20

Then b
(a, b)
( a, a ) = 0
(b, a ) c
0 (a, b) 1 a, b A
a
(a, b) 0 weak global preference
d
of a over b--over all criteria
(a, b) = 1 strong global preference Fuzzy outranking graph

Provides measures of (comparative)


intensity
21 22

E. PROMETHEE I:
x x Partial Preorder
a a
Let
Positive Negative aS+b IFF + (a ) > + (b) [Strong Preference]
outranking flow outranking flow aI +b IFF + ( a ) = + (b) [Positive Indifference]
(a ) = (a, x )
+
(a ) = ( x, a )

aS-b IFF ( a ) < (b) [Weak Preference]
xA xA

+
(a ) indicates how each action aI -b IFF ( a ) = (b) [Weak Indifference]
outranking others
Power or outranking character (in a global Gives partial relation (P I ,I I ,R)
sense)
23 24

4
a S+ b and aS-b

or 3 possible conclusions:
aP Ib IF aS+b and aI-b 1. a outranks b ( aP Ib )
or
aI + b and aS-b
Compare or higher power of a a combined
pairwise all with lower weakness of b
actions or
aI Ib IF aI +b and aI -b 2. a indifferent to b ( aI Ib )
alternatives
in set A positive and negative flows
aS+ b and bS a approximately equal
aR Ib IF or 3. a incomparable to b ( aRb )
bS+ a and aS-b
a good on one set of criteria on
which b weak, and vice versa
25 26

E. PROMETHEE II: E. EXAMPLELocation


Complete Preorder Problem
Use net outranking flow 6 possible locations to build a nuclear
power plant
(a ) = (a ) (a )
+
a1 Italy
aP b IFF ( a ) > (b)
II a2 Belgium
a3 Germany
aI IIb IFF (a ) = (b)
a4 UK
all actions comparable a5 Portugal
looks nicebut information gets lost a6 France
when finding balance of flow
27 28

6 criteria Evaluation Table


f1 manpower [minimize]
Criter- min a1 a2 a3 a4 a5 a6 Criter- Para-
f2 power (MW) [maximize] ion or ion meters
f3 construction costs ($106 US) max Type
f1 min 80 65 83 40 52 94 II q =10
[minimize] f2 max 90 58 60 80 72 96 III p =30
f4 annual maintenance costs (($106 US) f3 min 600 200 400 1000 600 700 V q=50 p=500
[minimize] f4 min 54 97 72 75 20 36 IV q=10 p=60
f5 min 8 1 4 7 3 5 I --
f5 ecologydisplacement of people f6 max 5 1 7 10 8 6 VI =5
[minimize]
f6 security [maximize] Assume wi = 0.167

29 30

5
(ai , a j ) Example: (a1, a2 )
f1 0 [a2Pa1]
a1 a2 a3 a4 a5 a6 +

f2 + 0.167 1 [a1Pa2]
a1 0 0.296 0.250 0.268 0.100 0.185 1.099 1.827 -0.728 f3 + 0 [a2Pa1]
a2 0.462 0 0.389 0.333 0.296 0.500 1.980 1.895 +0.085 f4 + 0.167 0.5 [a1Pa2]
a3 0.236 0.180 0 0.333 0.056 0.429 1.234 1.681 -0.447
a4 0.399 0.505 0.305 0 0.223 0.212 1.644 1.746 -0.102 f5 + 0 [a2Pa1]
a5 0.444 0.515 0.487 0.380 0 0.448 2.274 0.808 +1.466 f6 + 0.167 0.273 [a1Pa2]
a6 0.286 0.399 0.250 0.432 0.133 0 1.500 1.774 0.274
1.827 1.895 1.681 1.746 0.808 1.774
Sum 0.296

31 32

PROMETHEE I PROMETHEE II
+
a4S a6
a5S+a4 a4I -a6 a6S+ a1 1.466 -0.102 -0.447
a4 a6
a5S-a4 a6S- a1 a5 a4 a3
a5S+a3 a3S+ a1
a5 a3 a1
a5S-a3 a3S-a1
a2 a6 a1
a5S+ a2
a5S- a2
Assume transitivity 0.085 -0.274 -0.728
a2
applies

33 34

a5 selectedbut suppose we desire a


2nd choice
Cannot assume independencea2 not
necessarily best if a5 eliminated
Must redo analysis with a5 eliminated

_________________________
J. Brans, B. Mareshcal, P. Vincke, How to Select and Rank
Projects: The PROMETHEE Method, European Journal
of Operations Research, Vol. 24, 228-238, 1986.

35

Você também pode gostar