Parameters of the Genetic Algorithm: One of the
difficulties of genetic algorithms is finding the best internal
parameters in order to optimize speed and convergence.
These parameters are
• Size of the initial population
• Number of generations
• Crossover probability
• Mutation probability
• Rate of elitism.
The parameter values to be used are dependent on the type
of GA operators used. In order to be able to compare the
performance of the two crossover operators used, we tried to
keep the rest of the parameters the same. After quite a lot of
trial and error, the parameter values as presented in Table 2
appeared to be adequate. A rather low crossover probability
(0.1) together with a high mutation probability (1.0) yielded
the best results in the first series of experiments. This is
remarkable since it actually means that the two-point order based
crossover does not really works fine. The fact that this
way, eventually, optimal solutions are found yet seems to
be based on an almost less random search (caused by the
high mutation rate) in combination with the application of
an elitism strategy This was the reason to look for another cross-over operator,
the above-mentioned two-point bit equalizer crossover. In
the second series of experiments while applying this operator,
a more usual crossover probability (1.0) together with a more
normal, lower (but still quite high) mutation probability (0.5)
appeared to be best choices.
Initially, we used 50 generations of both GAs. For the GA
with the two-point bit equalizer crossover operator, this number
appeared to be large enough to guarantee convergence
to the same final solution in all experiments. However, this
number is not sufficient to guarantee convergence to the final
solution in case of using the GA with the two-point order based
crossover operator. We therefore also tried that GA
using 100 generations.