Optimization and Computational Fluid Dynamics - Department of ...
Optimization and Computational Fluid Dynamics - Department of ...
Optimization and Computational Fluid Dynamics - Department of ...
Create successful ePaper yourself
Turn your PDF publications into a flip-book with our unique Google optimized e-Paper software.
246 Marco Manzan, Enrico Nobile, Stefano Pieri <strong>and</strong> Francesco Pinto<br />
a) MOGA-II starts with an initial population P <strong>of</strong> size n <strong>and</strong> an empty<br />
elite set E = ∅<br />
b) For each generation, compute P ′ = P ∪ E<br />
c) If cardinality <strong>of</strong> P ′ is greater than cardinality <strong>of</strong> P, reduce P ′ r<strong>and</strong>omly<br />
removing exceeding points<br />
d) Generate P ′′ by applying MOGA algorithm to P ′<br />
e) Calculate P ′′ fitness <strong>and</strong> copy non-dominated designs to E<br />
f) Purge E from duplicated <strong>and</strong> dominated designs<br />
g) If cardinality <strong>of</strong> E is greater than cardinality <strong>of</strong> P r<strong>and</strong>omly shrink the<br />
set<br />
h) Update P with P ′′ <strong>and</strong>returntostep(b)<br />
2. NSGA-II. NSGA-II employs a fast non-dominated sorting procedure <strong>and</strong><br />
uses the crowding distance (which is an estimate <strong>of</strong> the density <strong>of</strong> solutions<br />
in the objective space) as a diversity preservation mechanism. Moreover,<br />
NSGA-II has an implementation <strong>of</strong> the crossover operator that allows the<br />
use <strong>of</strong> both continuous <strong>and</strong> discrete variables. The NSGA-II does not use<br />
an external memory as MOGA does. Its elitist mechanism consists <strong>of</strong> combining<br />
the best parents with the best <strong>of</strong>fspring obtained.<br />
8.7.3 Multi-Criteria Decision Making (MCDM)<br />
As already stated, it is impossible to find out a unique best solution in a<br />
multi-objective optimization process, but rather a whole group <strong>of</strong> designs<br />
that dominate the others: the Pareto front or Pareto optimal set.AllPareto<br />
optimal solutions can be regarded as equally desirable in a mathematical<br />
sense. But from an engineering point <strong>of</strong> view, the goal is a single solution to<br />
be put into practice at the end <strong>of</strong> an optimization. Hence, the need for a decision<br />
maker (DM) who is able to identify the most preferred one among the<br />
solutions. The decision maker is a person who is able to express preference<br />
information related to the conflicting objectives. Ranking between alternatives<br />
is a common <strong>and</strong> difficult task, especially when several solutions are<br />
available or when many objectives or decision makers are involved.<br />
Decisions have taken over a limited set <strong>of</strong> good alternatives mainly due<br />
to the experience <strong>and</strong> competence <strong>of</strong> the single DM. Therefore, the decision<br />
stage can be described as subjective <strong>and</strong> qualitative rather than objective <strong>and</strong><br />
quantitative. Multi-Criteria Decision Making (MCDM) refers to the solving<br />
<strong>of</strong> decision problems involving multiple <strong>and</strong> conflicting goals, coming up with<br />
a final solution that represents a good compromise that is acceptable to the<br />
entire team. As already underlined, when dealing with a multi-objective optimization,<br />
the decision making stage can be done in three different ways [53]: