Databases Reference
In-Depth Information
These references indicated that PSO-based ANN algorithms were
successful in evolving ANNs and achieved the generalization performance
comparable to or better than those of standard BP networks (BPNs)
or GA-based ANNs. However, they used PSO to evolve the parameters
(i.e., weights and bias) of ANNs without considering the optimization of
structure of the ANNs. Thus, the problem of designing a near optimal ANN
structure by using PSO for an application remains unsolved. However, this
is an important issue because the information processing capability of an
ANN is determined by its structure.
Although these researches have shown that PSO performs well for
global search because it is capable of quickly finding and exploring
promising regions in the search space, they are relatively inecient in
fine-tuning solutions.
94,95
Moreover, a potentially dangerous property in
PSO still exists: stagnation due to the lack of momentum, which makes it
impossible to arrive at the global optimum.
95
To avoid these drawbacks of
the basic PSO, some improvements such as the time-varying parameters
and random perturbation (e.g., velocity resetting)
95
have been proposed.
These improvements can enhance convergence of PSO toward the global
optimum, to find the optimum solution eciently. Yu
et al.
proposed
evolutionary ANN algorithm ESPNet based on an improved PSO/DPSO
with a self-adaptive ES. This integration of PSO and DPSO enables
the ANN to dynamically evolve its structure and adapt its parameters
simultaneously.
101
3.3.1.
Particle swarm optimization
The particle swarm algorithm is an optimization technique inspired by the
metaphor of social interaction observed among insects or animals.
94,95
The kind of social interaction modeled within a PSO is used to guide a
population of individuals (so called particles) moving towards the most
promising area of the search space. In a PSO algorithm, each particle is a
candidate solution equivalent to a point in a
d
-dimensional space, so the
i
th particle can be represented as
x
i
=(
x
i
1
,x
i
2
,...,x
id
). Each particle
“flies” through the search space, depending on two important factors,
p
i
=(
p
i
1
,p
i
2
,...,p
id
), the best position the current particle has found
so far; and
p
g
=(
p
g
1
,p
g
2
,...,p
gd
), the global best position identified from
the entire population (or within a neighbourhood).
