Forward Backward Transformation - Differential Evolution: A Topic for Global Permutation-Based Combinatorial Optimization - page 62

6 10 14 19 20

10 16 18 23 26

13 19 23 24 29

17 23 28 31 32

Makespan

Fig. 3.14. Accumulated time Gantt Chart

Table 3.27. DDE FSS operational values

Parameter Values

NP 150

CR 0.9

F 0.3

Table 3.28. Comparison of 10 DE-strategies using the 10x25 problem data set

Strategy

7 a 1 2 3 4 5 6 8 9 10

Makespan 211.8 209.2 212.2 212.4 208.6 210.6 207.8 212.4 210 207.2
Total tardiness 3001.8 3034.6 3021.4 3089.2 3008 2987.8 2936.4 3034.2 2982.8 2990.6
Mean flowtime 105.75 105.11 105.52 107.71 104.68 103.03 103.17 105.32 104.7 104.16
a Strategy 7 is the best.

In order to operate, the first phase is to obtain the optimal tuning parameters. All
parameters were obtained empirically. The values are given in Table 3.27.
The second phase was to obtain the optimal strategy. Through experience in solving
these problems, it became evidently clear that not all the strategies behaved similarly,
hence the need to isolate the most promising one from the ten different.
An arbitrary problem of average difficulty was selected, in this case the 10x25 job
problem, and using the selected parameters, ten iterations were done. The average val-
ues are presented in Table 3.28. Using the multi-objective function of makespan, tardi-
ness and flowtime, Strategy 7 was selected as the optimal.