Professional Documents
Culture Documents
NSGA2 Based Route Optimization
NSGA2 Based Route Optimization
CHAPTER 4
Non-dominated Crowding
sorting distance sorting
P t+1
F1
Pt F2
F3
Qt Rejected
Rt
Step 3: For each non-dominated solution, assign a fitness (rank) equal to its
non-domination level (1 is the best level, 2 is the next best level,
and so on).
Step 5: From the first generation onwards, creation of each new generation
constitutes the following steps:
Start
Initialize population
(size = Z)
Evaluation of objective
functions
Rank population
Selection
Crossover
Mutation
Evaluation of objective
functions
Yes
Report solution and
stop
f1
0
i–1
i
i+1
l
f2
Step 2: For each objective function fm, m =1, 2,..., M, sort the set in worse
order of fm or, find the sorted indices vector, Fm = sort (fm, >).
– 1), assign:
m m
f m( F )
v 1
fm( F v 1)
dF m dF m (4.1)
v v 1
f mmax fmmin
118
m
fm( F v 1)
is the mth objective function value of the (v+ 1)th individual in
the set F r
m
fm( F v 1)
is the mth objective function value of the (v–1)th individual in
the set F r
generation. The individual with the lowest front number is selected if the two
individuals are from different fronts. The individual with the highest
crowding distance is selected if they are from the same front. A higher fitness
is assigned to individuals located on a sparsely populated part of the front. In
each iteration, the existing Z parents generate Z new offspring. Both parents
and offspring compete with each other for inclusion in the next iteration.
4.3.8 Crossover
PMX scheme is used for crossover operation. The reason for doing
so is given in Section 3.3.4.2 of the previous chapter.
4.3.9 Mutation
receiver was the twentieth node since that would give the largest number of
possible paths in the network. Table 4.1 shows the set of control parameters
selected after conducting several experiments.
Maximum
Number Population ET RO
Solutions
of nodes size (in Sec.) (%)
obtained
20 40 23 3.5 100
40 80 32 4.6 100
60 120 45 5.9 100
80 160 65 6.8 100
100 200 98 8.21 100
From the results, it is observed that, when the size of the network
was increased, the number of solutions found by the NSGA-II also increased.
Maximum
Number Population ET RO
Solutions
of nodes size (in Sec.) (%)
obtained
200 200 152 8.42 100
300 300 284 8.89 100
400 400 352 9.13 99
500 500 407 9.45 98
600 600 415 9.7 98
700 700 424 10 98
800 800 482 10.4 97
900 900 484 10.8 97
1000 1000 492 11.3 95
size networks that could be noted from Table 4.2. Even up to 300 nodes, the
RO obtained by NSGA –II was 100%, i.e. in all the simulation runs carried
out, the Pareto-optimal solutions were obtained. When the network size
becomes larger, the RO tends to get reduced. Obviously, for the 1000 node
network case, it is only 95%. Figure 4.4 shows the variation of RO with
increase in number of nodes.
NSGA-II
101
100
99
Route Optimality
98
97
96
95
94
93
92
200 400 600 800 1000
Number of nodes
Tables 4.2 and 4.3 give the ET needed by the CPU to identify the
Pareto-optimal set. Its variations against the number of nodes are presented in
Figures 4.5 and 4.6.
123
NSGA-II
10
0
20 40 60 80 100 120
Number of nodes
NSGA-II
14
13
12
Execution Time (seconds)
11
10
5
200 400 600 800 1000
Number of nodes
When the number of nodes in the network was 20, the ET was 3.5
seconds and when the size of the network was increased, the ET also
increased. For a 100 node network the ET was found to be 8.21 seconds.
When increasing the number the nodes further, it was observed that the ET
also increased. When the maximum network size was taken to be 1000, the
maximum ET was found to be 11.3 seconds. The ET for various network
sizes is plotted in Figure 4.5 and Figure 4.6.
NSGA-II
600
500
Number of Solutions
400
300
200
100
Number of nodes
Maximum
Number ET RO
Solutions
of nodes (in Sec.) (%)
obtained
20 32 9.7 100
40 41 11.4 100
60 53 13.9 100
80 65 15.3 100
100 74 16.2 99
126
NSGA-II
Route Optimality 100
99.5
99
98.5
20 40 60 80 100
Number of nodes
From the results that are tabulated in Table 4.4, it is clear that the
RO obtained by NSGA-II was 100% when the size of the network was varied
from 20 to 80 nodes.
When the number of nodes in the network was increased from 100
to 1000 the RO decreased from 99% to 95%. The worst route optimality
obtained was 95% when the size of the network had maximum number of
nodes considered in this thesis. The variation of RO with increase in number
of nodes is shown in Figure 4.8 and Figure 4.9 for the two groups, smaller and
larger sizes separately.
Maximum
Number ET RO
Solutions
of nodes (in Sec.) (%)
obtained
200 159 17.5 99
300 223 18.25 98
400 294 19.7 98
500 340 21.23 97
600 400 22.68 97
700 431 24.01 97
800 482 25.64 96
900 495 25.72 95
1000 493 25.84 95
NSGA-II
100
99
Route Optimality
98
97
96
95
94
93
200 300 400 500 600 700 800 900 1000
Number of nodes
NSGA-II
20
15
Execution Time (seconds)
10
0
20 40 60 80 100 120
Number of nodes
The ET for various network sizes when the number of nodes in the
network was varied from 20 to 100 is plotted in Figure 4.10. Similarly, the ET
for various network sizes when the number of nodes in the network was
varied from 100 to 1000 is plotted in Figure 4.11. From Figures 4.10 and
4.11, it is observed that when the number of nodes in the network is
increased, the ET increases almost exponentially. The maximum ET was
25.84 seconds when the number of nodes in the network was made 1000.
NSGA-II
30
20
15
10
0
200 400 600 800 1000
Number of nodes
NSGA-II
80
Number of Solutions
60
40
20
0
20 40 60 80 100
Number of nodes
NSGA-II
600
Number of Solutions
500
400
300
200
100
0
200 300 400 500 600 700 800 900 1000
Number of nodes
Table 4.6 Execution Time variation of NSGA-II with different group size
problems
NSGA-II
60
Group size 25%
Group size 35%
50 Group size 45%
Execution time (seconds)
30
20
10
4.7 CONCLUSION