Information Technology Reference
In-Depth Information
1
Start
2
t:=0;
3
Initialize
P
(
t
);
4
Repeat
5
For
k=1
To
λ
Do
6
Choose
ρ
parents from
P
(
t
)
7
σ
k
:=recombination strategy variables;
8
x
k
:=recombination objective variables;
9
σ
k
:=mutation strategy variables;
10
x
k
:=mutation objective variables;
11
F
k
:=
f
(
x
k
);
12
Next
13
O
(
t
):=
{
(
x
1
,σ
1
,F
1
)
,...,
(
x
λ
,σ
λ
,F
λ
)
}
;
14
Select
P
(
t
+1) from
O
(
t
)
∪P
(
t
)inthecaseof(
μ
+
λ
)-selection,
15
or from
O
(
t
)inthecaseof(
μ, λ
)-selection;
16
t
:=
t
+1;
17
Until
termination condition
18
End
Fig. 2.3.
Pseudo code of a (
μ/ρ
+
,λ
)-ES
After
λ
individuals are produced, the best
μ
individuals are selected as par-
ents for the next generation exclusively from the offspring population in case
of
comma selection
or from the offspring and the previous parental popula-
tion in case of the
plus selection
scheme. Figure 2.3 shows the pseudocode of a
(
μ/ρ
+
,λ
)-ES.
2.2.2
The (
μ, κ, λ, ρ
)-ES
The comma selection limits the maximal lifespan of an individual to one gener-
ation, while an individual has an unlimited lifespan under plus selection. This is
not necessarily the case. The parameter
κ>
0 introduces a maximal number of
generations an individual is allowed to survive. After
κ
generations the individ-
ual will be replaced, even if no better or equal solution exists. This variant of ES
is called (
μ, κ, λ, ρ
)-ES. The TSES in chapter 7 makes use of the
κ
-parameter.
2.3 Practical Guidelines for Evolutionary Algorithms
In the following we summarize some practical EA wisdoms, which haven proven
their worth in recent projects.
•
The
representation
should be chosen wisely as on the one hand it determines
the size of the search space. On the other hand, the representation has a
direct influence on the impact of genetic operators.
•
The
population sizes
control the
diversity
. Small population sizes might be
ecient, but often have to be paid with a decrease in quality of the results
(eciency vs. accuracy).
Search WWH ::
Custom Search