Information Technology Reference
In-Depth Information
Table 7.4.
Above: Experimental analysis of the TSES on problem g01. The standard
DP method and the (8+8,13+87)-TSES show bad results, an increase of population
sizes is necessary. All other experiments were more successful, the optimum could be
reached in almost every run. Below: the TSES on problem g12. The TSES approximates
the optimum
−
1
.
0 in most of the experiments with sucient accuracy. In some runs
of the (8 + 8
,
13 + 87)-TSES with fixed starting points a premature termination was
observed.
TSES
κ
best
avg
worst
dev
σ
ffc/cfc
g01
(8 + 8
,
13 + 87)
200
−
15
.
0
−
14
.
8008668542
−
12
.
46168543 5.4E-2
1.4E-11 360 599
(20 + 20
,
25 + 200) 50
−
15
.
0
−
15
.
0
−
15
.
0
9.1E-15 2.7E-14 243 920
(20 + 20
,
25 + 200)
∗
50
−
15
.
0
−
14
.
8958084010
−
12
.
44021510 4.0E-2
2.9E-14 277 823
(40 + 40
,
50 + 400) 50
−
15
.
0
−
15
.
0
−
15
.
0
5.3E-15 1.8E-14 411 732
(40 + 40
,
50 + 400)
∗
50
−
15
.
0
−
15
.
0
−
15
.
0
5.5E-15 1.7E-14 468 580
(15
,
100)-DP
0
−
15
.
0
−
14
.
1734734986
−
11
.
58255662 1
.
106
5.1E-11 143 187
g12
(8 + 8
,
15 + 85)
50
−
1
.
0
−
1
.
0
−
1
.
0
0
.
0
5.5E-8
12 305
(8 + 8
,
15 + 85)
∗
50
−
1
.
0
−
1
.
0
−
1
.
0
0
.
0
5.9E-8
11 722
(8 + 8
,
13 + 87)
50
−
1
.
0
−
0
.
9999437499
−
0
.
994374999 5.6E-5
1.7E-7
14 534
(8 + 8
,
13 + 87)
∗
50
−
1
.
0
−
1
.
0
−
1
.
0
0
.
0
6.9E-8
12 720
(8 + 8
,
10 + 90)
50
−
1
.
0
−
1
.
0
−
1
.
0
7.4E-16 1.5E-6
18 938
(8 + 8
,
10 + 90)
∗
50
−
1
.
0
−
1
.
0
−
1
.
0
1.6E-16 1.4E-6
17 237
(15
,
100)-DP
−
1
.
0
−
1
.
0
−
1
.
0
0
2.9E-13 5.7E-5
20 318
experiments on problem g01. Problem g01 exhibits a quadratic objective func-
tion and nine linear inequality constraints. The experiments marked with a star
make use of randomized starting individuals. The other tests use a fixed initial-
ization. The (8+8,13+87)-TSES was not at all able to approximate the opti-
mum. But a modification of the sex ratios could change the situation. Both, a
(20+20,25+200)- and a (40+40,50+400)-TSES, were able to find the optimum
with arbitrary accuracy in almost every run. Only the (20+20,25+200)-TSES
with randomized starting point suffered from premature step size reduction be-
fore reaching the optimum. In comparison to death penalty a significant im-
provement could be observed.
Experimental settings
Population model
(
μ
o
+
μ
c
,λ
o
+
λ
c
)
Mutation type standard,
n
σ
=
N
,
τ
0
=(
√
2
n
)
−
1
and
τ
1
=(
2
√
n
)
−
1
Crossover type intermediate,
ρ
=2
Selection type comma,
κ
, two-step-selection sex o
Initialization
σ
i
=
|x
(0)
−x
∗
|
N
Termination fitness stagnation,
θ
=10
−
12
Constraint handling
TSES
Runs
25
Search WWH ::
Custom Search