Databases Reference
In-Depth Information
Holden and Freitas
104
proposed hybridization of PSO and ACO
algorithms for hierarchical classification. They applied it to the functional
classification of enzymes and obtained very promising results.
PSO was combined with differential evolution (DE) by Hendtlass.
105
He
obtained mixed results for this hybridization. On one multimodal problem
the hybridized algorithm performed better than PSO or DE. But PSO was
found to be faster and robust than DE and hybrid models. Subsequently
Zhang and Xie
106
used hybridization of PSO and DE and reported to obtain
better results.
Poli
et al.
107,108
proposed hybridization of PSO with genetic
programming (GP). For the control of the particle movement GP was used
to evolve new laws. This method provided better result than standard PSO
methods.
9.3.4.
PSOs with diversity control
It has been reported by different researchers that the swarm has a tendency
of converging prematurely to local optima. Different approaches have
been suggested to overcome the premature convergence as the swarm
concentrates on a single optimum.
To help the PSO to attain more diversity and became less vulnerable
to local minima, Loovbjerg
109
proposed critically self organized PSO.
In his method if two particles became close to one another, a variable
called the “critical value” is incremented. When the variable reaches
the criticality threshold, one of the particles is allowed to relocate
itself.
Some researchers have tried to diversify the particles' clustering too
closely in one region of the search space. Blackwell and Bentley
110
suggested
a method called the collision-avoiding swarms which achieves diversity by
reducing the attraction of the swarm center.
The “spatially extended” particles were proposed by Krink
et al.
,
111
where each particle was considered to be surrounded by a sphere of some
radius and if such a particle collides with another particle, it bounces off.
A negative entropy value is added to the particle swarm by Xie
et al.
112
to discourage excessively rapid convergence towards a poor quality local
optimum. Considering different conditions they have weighed the velocity,
location of a particle with some random value to obtain dissipative particle
swarm.
