Information Technology Reference
In-Depth Information
subproblem. For each of the three data sets described below 20 instances were
randomly generated based on distributions from real-world instances provided
by the industrial partner.
Set A (small-size instances): 10 nodes, 20 trailers, 10 products.
Set B (medium-size instances): 15 nodes, 50 trailers, 40 products.
Set C (large-size instances): 25 nodes, 150 trailers, 300 products.
Small-size Instances
80.0%
70.0%
60.0%
E
50.0%
= 0.00
= 0.25
= 0.50
= 0.75
= 1.00
40.0%
E
30.0%
E
20.0%
E
10.0%
E
0.0%
D
Medium-size Instances
25.0%
20.0%
E
= 0.00
= 0.25
= 0.50
= 0.75
= 1.00
15.0%
E
10.0%
E
E
5.0%
E
0.0%
D
Large-size Instances
14.0%
12.0%
10.0%
E
= 0.00
= 0.25
= 0.50
= 0.75
= 1.00
8.0%
E
6.0%
E
4.0%
E
2.0%
E
0.0%
D
Fig. 2. Algorithm behavior as a function of α and β
Fine-Tuning of Parameters
The first experiment aims at finding appropriate values for algorithmic param-
eters α and β . To this end, we run a full two-factorial design. Parameter β was
fixed at values in the [0.00, 1.00] range with increments of 0.25. Parameter α
was fixed at values in the range [0.00, 0.80], [0.00, 0.50], and [0.00, 0.20] for the
small-, medium-, and large-size instances, respectively, with increments of 0.01.
Search WWH ::




Custom Search