Self-Adaptive Inversion Mutation - Self-Adaptive Heuristics for Evolutionary Computation - page 84

25

SA-INV 5-1
SA-INV 10-2
SA-INV 20-5

20

15

10

5

0

0 10 20 30 40 50 60 70 80 90 100

generations

650000

SA-INV 5-1
SA-INV 10-2
INV 1
INV 10

600000

550000

500000

450000

400000

350000

300000

250000

200000

0 10 20 30 40 50 60 70 80 90 100

generations

Fig. 5.4. SA-INV on problem bier127 . Upper part: Development of k averaged over
25 runs. For all SA-INV variants the mutation strength increases in the first gener-
ations. Lower part: Fitness development of SA-INV and INV averaged over 25 runs.
SA-INV 5 − 1 is the fastest variant until generation t ≈ 80. INV 10 is very fast at the
beginning, but suffers from fitness stagnation after generation t ≈ 20.

Table 5.3. Experimental analysis of SA-INV on problem gr666 with 666 cities .All
SA-INV variants are faster than the standard INV operators with fixed k .

SA-INV 5 − 1 SA-INV 10 − 2 SA-INV 20 − 5 INV 1 INV 10

best 3074023 3062252 3023938 3198055 3270915

median 3150807 3137811 3189521 3340901 3353568

worst 3382585 3497183 3281412 3521550 3458274

mean 3180473.08 3159244.16 3180724.16 3347929.96 3355142.4

dev 84830.42 87259.16 73612.26 86277.86 47431.11