Scheduling in Manufacturing Systems: The Ant Colony Approach
Read this section. It makes the case that production systems are similar to ant colonies because ants continue to learn from one another and find ways to become more efficient at their task at hand. As you read, pay special attention to some of the scheduling problems.
5. Comparing ACO algorithm and neural algorithm
For multiple criteria optimization in the following tests comparisons were made of compromise solutions for ACO algorithm with the results of neural algorithm. Optimization criteria were: time, cost, and power consumption. Additional requirements and constraints were adopted: maximum number of processors – 5, maximal cost – 3, maximal time – 25.
|
Number of tasks |
Ant colony |
Neural |
||||
|
Cost |
Time |
Power consumption |
Cost |
Time |
Power consumption |
|
|
5 |
1,75 |
6,75 |
9,26 |
1,00 |
3,90 |
4,39 |
|
10 |
1,50 |
6,20 |
35,47 |
1,50 |
8,50 |
11,61 |
|
15 |
2,75 |
18,00 |
22,96 |
2,00 |
16,00 |
17,85 |
|
20 |
1,75 |
12,83 |
35,45 |
2,00 |
22,50 |
20,31 |
|
25 |
2,00 |
14,50 |
51,25 |
2,00 |
22,00 |
28,93 |
|
30 |
2,75 |
16,90 |
63,58 |
2,50 |
23,00 |
35,01 |
|
35 |
2,00 |
18,00 |
78,30 |
2,50 |
24,67 |
36,12 |
|
40 |
2,75 |
17,75 |
104,68 |
2,50 |
17,00 |
72,52 |
|
45 |
2,25 |
21,75 |
99,50 |
2,50 |
18,67 |
79,02 |
|
50 |
2,25 |
23,88 |
113,26 |
2,50 |
21,00 |
88,57 |
|
55 |
2,50 |
25,00 |
164,58 |
2,50 |
22,50 |
95,33 |
Table 5.
Comparison of Ant
Colony and neural for minimization of time, cost, and power consumption.
Results were illustrated on the following charts – Chart: 30, 31, and 32.
When comparing solutions obtained by the algorithms one cannot provide an unequivocal answer which of the optimization methods is better. Greater influence on the quality of offered solutions has the algorithm itself, especially its exploration capacity of admissible solutions space. When analyzing the graphs of interdependence between cost and task number, it appears that neural algorithm is more stable i.e. attempts to maintain low cost, despite an increase in the number of tasks. This results in worse task performance time what is very visible on the graph where time is contingent on the number of tasks. From power consumption analysis it is evident that ACO algorithm solutions are more beneficial.
Chart 30.
Influence of number tasks on cost –
minimization of time, cost, and power consumption.
Chart 31.
Influence of number of tasks on time –
minimization of time, cost, and power consumption.
Chart 32.
Influence of number of tasks on power
consumption – minimization of time, cost, and power consumption.
Additional requirements and constraints were adopted: maximum number of processors: 5, maximal cost: 8, maximal time: 50.
Results were illustrated in the following charts - Chart: 33, 34, and 35.
Chart 33.
Influence of number of tasks on cost –
minimization of time, cost, and power consumption with of cost of memory.
Chart 34.
Influence of number of tasks on time –
minimization of time, cost, and power consumption with of cost of memory.
|
Number of tasks |
Ant colony |
Neural |
||||
|
Cost |
Time |
Power consumption |
Cost |
Time |
Power consumption |
|
|
10 |
6,50 |
2,00 |
37,99 |
4,50 |
6,00 |
7,52 |
|
20 |
1,50 |
18,50 |
33,15 |
4,00 |
11,00 |
19,07 |
|
30 |
5,90 |
23,00 |
82,41 |
5,00 |
14,00 |
30,98 |
|
40 |
7,00 |
23,00 |
121,56 |
5,00 |
18,00 |
37,33 |
|
50 |
4,25 |
16,20 |
186,05 |
5,00 |
21,00 |
49,99 |
|
60 |
2,50 |
32,00 |
175,24 |
5,00 |
25,00 |
60,20 |
|
70 |
2,50 |
38,00 |
167,59 |
5,00 |
29,00 |
69,35 |
|
80 |
3,25 |
37,00 |
183,67 |
5,00 |
32,00 |
79,19 |
|
90 |
4,25 |
28,60 |
328,73 |
5,00 |
36,00 |
98,39 |
|
100 |
6,75 |
30,33 |
336,36 |
5,50 |
39,00 |
101,62 |
|
110 |
4,25 |
41,80 |
435,77 |
5,00 |
43,00 |
115,53 |
Table 6.
Comparison of Ant
colony and neural for minimization of time, cost, and power consumption.
Chart 35.
Influence of number of tasks on power
consumption – minimization of time, cost, and power consumption with memory cost