Progressively Interactive Evolutionary Multi-Objective Optimization ...
Progressively Interactive Evolutionary Multi-Objective Optimization ...
Progressively Interactive Evolutionary Multi-Objective Optimization ...
You also want an ePaper? Increase the reach of your titles
YUMPU automatically turns print PDFs into web optimized ePapers that Google loves.
the lower level, solution y can be better than solution x in the upper level. Due<br />
to these discrepancies, these problems will cause a conflict in converging to the<br />
appropriatePareto-optimalfrontinbothlowerandupperleveloptimizationtasks.<br />
6. Extension to higher level optimizationproblems is possible. Although our focus here is<br />
for bilevel problems only, test problems scalable to three or higher levels would<br />
be interesting, as there may exist some practical problems formulated in three or<br />
higher levels. On the other hand, it will alsobe ideal tohave bilevel test problems<br />
whichwilldegeneratetochallengingsingleleveltestproblems,ifasingleobjective<br />
function is chosenfor eachlevel.<br />
7. Differentlowerlevelproblemsmaycontributedifferentlytotheupperlevelfrontintermsof<br />
theirextentofrepresentativesolutionson theupperlevel Pareto-optimalfront. Thesetest<br />
problemswilltestanalgorithm’sabilitytoemphasisdifferentlowerlevelproblems<br />
differentlyinordertofind awell-distributedsetof Pareto-optimalsolutions atthe<br />
upperlevel.<br />
8. Test problems must include constraints at both levels. This will allow algorithms to be<br />
tested for their ability to handle constraints in both lower and upper level optimizationproblems.<br />
Different principles are possible to construct test problems following the above<br />
guidelines. Here, we present a generalized version of a recently proposed procedure<br />
(Deband Sinha,2009a).<br />
4.5.1 A<strong>Multi</strong>-<strong>Objective</strong>BilevelTestProblemConstructionProcedure<br />
We suggest atest problemconstruction procedureforabilevel problemhaving M and<br />
m objectives in the upper and lower level, respectively. The procedure needs at most<br />
threefunctional forms andis describedbelow:<br />
Step1: First, a parametric trade-off function ΦU : R M−1 → R M which determines a<br />
trade-off frontier (v1(u), . . . , vM (u)) on the F-space as a function of (M − 1) parameters<br />
u (can be considered as a subset of xu) is chosen. Figure 3 shows such a<br />
v1-v2 relationship on atwo-objective bilevel problem.<br />
Step2: Next, for every point v on the ΦU-frontier, a (M − 1)-dimensional envelope<br />
(U1(t), . . . , UM(t)) v ) on the F-space as a function of t (having (M − 1)<br />
parameters) is chosen. The non-dominated part of the agglomerate envelope<br />
∪v ∪ t<br />
(v1(u) + U1(t) v ), . . . , (vM (u) + UM (t) v ) constitutes the overall upper<br />
levelPareto-optimalfront. Figure3indicatesthis upperlevelPareto-optimalfront<br />
andsomespecificPareto-optimalpoints(markedwithbiggercircles)derivedfrom<br />
specific v-vectors.<br />
Step3: Next, for every point v on the ΦU-frontier, a mapping function ΦL : R M−1 →<br />
R m−1 which maps every v-point from the U-frontier to the lower level Paretooptimalfront<br />
(f ∗ 1 (s), . . . , f ∗ m (s))v ischosen. Here, sisa(m−1)-dimensionalvector<br />
andcanbeconsideredasasubsetof xl. Figure3showsthismappingTheenvelope<br />
A ′ C ′ B ′ (a circle in the figure) is mapped to the lower level Pareto-optimal frontier<br />
ACB(inlet figureon top).<br />
Step4: After these three functions are defined, the lower level problem can be constructed<br />
by using a bottom-up procedureadopted in Deb et al. (2005)through additional<br />
terms arising from other lower level decision variables: fj(xl) = f ∗ j (s) +<br />
83