Databases Reference
In-Depth Information
A random number
r
is generated for each bit and compared it to
s
(
x
i,d
). If
r
was less that the threshold, then
x
i,d
was interpreted as 1,
otherwise as 0.
Agrafiotis and Cedeno
78
performedfeatureselec ioninpa tern
matching task by using the particle locations as probabilities to select
features. Basing on the location value of the particle each feature was
assigned a slice of a roulette wheel. Depending on the selection of feature,
the values are discretized to 0, 1.
Mohan and Al-Kazemi
94
suggested different methods for implementing
particle swarm on binary space. They have suggested a method called
“regulated discrete particle swarm,” which performs well on a test problem.
In Pampara
et al.
95
encoded each particle with small number of
coecients of a trigonometric model (angle modulation) which was then
run to generate bit strings.
Clerc
96,97
: Moraglio
et al.
98
have extending PSO to more complex
combinatorial search spaces and observed some progress. However it is
dicult to predict if PSO will be a better choice for such combinatorial
search spaces.
9.3.3.
Hybrids and adaptive particle swarms
Different researchers have tried to utilize the information from the
environment for fine tuning the PSO parameters. Evolutionary computation
and other techniques have been followed for the purpose.
Angeline
42
hybridized particle swarms in his model. He applied
selection to the particles, then the “good” particles were reproduced and
mutated, but the “bad” particles were eliminated. He obtained improved
results with this modification.
Evolutionary strategies concept was used by Miranda and Fonseca
92
to improve the performance of PSO. They modified the particle values by
adding random values distributed around a mean of zero; the variance of the
distribution is evolved along with function parameters. They used Gaussian
random values to perturb
χ
,
φ
1
,and
φ
2
,aswellasthepositionofthe
neighborhood best, but not the individual best by using selection to adapt
the variance. The evolutionary self-adapting particle swarm optimization
method has shown excellent performance in comparison to some standard
particle swarm methods. They have used it for the manufacture of optical
filters and for optimization of power systems.
