Progressively Interactive Evolutionary Multi-Objective Optimization ...
Progressively Interactive Evolutionary Multi-Objective Optimization ...
Progressively Interactive Evolutionary Multi-Objective Optimization ...
Create successful ePaper yourself
Turn your PDF publications into a flip-book with our unique Google optimized e-Paper software.
is at most 10% of the maximum difference in value functions<br />
between ≻-class of points.<br />
A little thought will reveal that the above optimization<br />
problem attempts to find a value function for which the<br />
minimum difference in the value function values between<br />
the ordered pairs of points is maximum. Considering all the<br />
expressions, we have the following optimization problem:<br />
Maximize ǫ,<br />
subject to Sm(Pi) ≥ 0, i = 1, 2, . . . , η, and m = 1, 2,<br />
S2(Pi) + k2S1(Pi) ≥ 0, i = 1, 2, . . . , η,<br />
k1S2(Pi) + S1(Pi) ≥ 0, i = 1, 2, . . . , η,<br />
V (Pi) − V (Pj) ≥ ǫ, for all (i, j) pairs<br />
satisfying Pi ≻ Pj,<br />
|V (Pi) − V (Pj)| ≤ δV , for all (i, j) pairs<br />
satisfying Pi ≡ Pj.<br />
(4)<br />
Figure 2 considers five (η = 5) hypothetical points (P1 =<br />
(3.5, 3.7), P2 = (2.6, 4.0), P3 = (5.9, 2.2), P4 = (0.0, 6.0),<br />
and P5 = (15.0, 0.5)) and a complete ranking of the points (P1<br />
being best and P5 being worst). Due to a complete ranking, we<br />
do not have the fourth constraint set. The solution to the above<br />
optimization problem results in a value function, the contours<br />
(iso-utility curves) of which are drawn in the figure. The value<br />
function obtained after the optimization is as follows:<br />
V (f1, f2) = (f1 + 4.3229)(f2 + 0.9401).<br />
The asymptotes of this value function are parallel to f1 and<br />
f2 axes. The optimized value of ǫ is 2.0991. It is interesting to<br />
note the preference order and other restrictions are maintained<br />
by the obtained value function.<br />
f 2<br />
11<br />
10<br />
9<br />
8<br />
7<br />
6<br />
5<br />
4<br />
3<br />
2<br />
1<br />
0<br />
P4<br />
P2<br />
P1<br />
P3<br />
0 2 4 6 8 10 12 14 16<br />
f<br />
1<br />
Fig. 2. Value function found by optimization.<br />
Interestingly, if the DM provides a different preference<br />
information: P1 is preferred over P2, P2 is preferred over P3<br />
and no preference exists among P3, P4 and P5, a different<br />
value function will be obtained. We re-optimize the resulting<br />
problem with the above preference information on the same<br />
set of five points used in Figure 2 and obtain the following<br />
value function:<br />
V (f1, f2) = (f1 + 5.9355)(f2 + 1.6613).<br />
P5<br />
34<br />
Figure 3 shows the corresponding value function contours.<br />
The contour makes a clear distinction between solutions in<br />
pairs P1-P2 and P2-P3 (within an optimized value ǫ), however,<br />
there is no distinction among P3, P4 and P5 (with 0.1ǫ), to<br />
establish the given preference structure. Since a value function<br />
maintaining a difference (ǫ) between points in pairs P1-P2 and<br />
P2-P3 is needed and a maximum gap of 10% of ǫ is needed,<br />
a somewhat greater ǫ value to that found in the previous case<br />
is obtained here. The optimized ǫ value is found to be 2.2645<br />
in this case.<br />
Fig. 3. Revised value function with a different preference information.<br />
2) Polynomial Value Function for M <strong>Objective</strong>s: The above<br />
suggested methodology can be applied to any number of<br />
objectives. For a general M objective problem the value<br />
function can be written as follows:<br />
V (f) = (f1 + k11f2 + k12f3 + . . . + k 1(M−1)fM + l1)×<br />
(f2 + k21f3 + k22f4 + . . . + k 2(M−1)f1 + l2)×<br />
. . .<br />
(fM + kM1f1 + kM2f4 + . . . + k M(M−1)fM−1 + lM )<br />
(5)<br />
The above value function can be expressed more elegantly as<br />
follows:<br />
⎛<br />
M M<br />
V (f) = ⎝<br />
i=1<br />
j=1<br />
Kijfj + K i(M+1)<br />
⎞<br />
<br />
⎠ . (6)<br />
Since each term in the value function can be normalized, we<br />
can introduce an additional constraint M<br />
j=1 Kij = 1 for each<br />
term denoted by i. As discussed below, Kij ≥ 0 for j ≤ M<br />
and for each i, however K i(M+1) can take any sign. In the<br />
remainder of the paper, we follow the value function definition<br />
given in equation 5.<br />
In the formulation it should be noted that the subscripts<br />
of the objective functions change in a cyclic manner as we<br />
move from one product term to the next. The number of<br />
parameters in the value function is M 2 . The optimization<br />
problem formulation for the value function suggested above<br />
contains M 2 + 1 variables (kij and li). The variable ǫ is to be<br />
maximized. The second set of constraints (strictly increasing<br />
property of V ) will introduce non-linearity. To avoid this,<br />
we simplify the above constraints by restricting the strictly<br />
increasing property of each term Sk, instead of V itself. The