Databases Reference
In-Depth Information
Shi and Eberhart
42
have found a significant improvement in the
performance of PSO with the linearly decreasing inertia weight over
the generations, time-varying inertia weight (TVIW) which is given in
Equation (5.8):
w
=
w
2
+
maxiter
iter
maxiter
−
∗
(
w
1
−
w
2
)
,
(5.8)
where
w
1
and
w
2
are the higher and lower inertia weight values and the
values of
w
will decrease from
w
1
to
w
2
.
iter
is the current iteration (or
generation) and
maxiter
is the maximum number of iteration (or total
number of generation).
Then, Ratnaweera and Halgamuge
44
introduced a time varying
acceleration co-ecient (TVAC), which reduces the cognitive component,
c
1
and increases the social component,
c
2
of acceleration co-ecient with
time. With a large value of
c
1
and a small value of
c
2
at the beginning,
particles are allowed to move around the search space, instead of moving
toward pbest. A small value of
c
1
and a large value of
c
2
allow the particles
converge to the global optima in the latter part of the optimization. The
TVAC is given in Equations (5.9) and (5.10):
maxiter
+
c
1
f
iter
maxiter
−
c
1
=(
c
1
i
−
c
1
f
)
∗
(5.9)
maxiter
+
c
2
f
iter
maxiter
−
c
2
=(
c
2
i
−
c
2
f
)
∗
(5.10)
where
c
1
i
and
c
2
i
are the initial values of the acceleration coecient
c
1
and
c
2
and
c
1
f
and
c
2
f
are the final values of the acceleration co-ecient
c
1
and
c
2
, respectively.
Thus far we have discussed PSO for continuous space, however, many
optimization problems including the problem to be solved in this chapter are
set in a space featuring discrete, qualitative distinctions between variables
and between levels of variables.
5.4. HAPSO for Learnable Bayesian Classifier in IDS
5.4.1.
Adaptive PSO
In the standard PSO method, the inertia weight is made constant for all
the particles in a single simulation, but the most important parameter that
