Information Technology Reference
In-Depth Information
Step 2.
Initialize FCM with
c
and
V
0
, execute FCM clustering and decide
the final result of clustering: the centers of clusters
V
and the fuzzy partition
matrix
U
.
Step 3.
Obtain the parameters (
a
ij
,b
ij
,c
ij
) and the candidate rule set
GP
=
{
by LSE. Then let
RB
i
be an empty set,
i
=1
,...,c
,
j
=1
,...,p
+1.
Step 4.
Select the rule
R
i
from GP to be adjusted which is the most consistent
with the rules in
RB
i
.
Step 5.
Optimize the parameters of
R
i
by (1+1) ES.
Step 6.
Remove
R
i
from GP.
Step 7.
Judge: if GP is not an empty set, let
i
=
i
+1and
RB
i
+1
=
RB
i
R
i
}
∪
R
i
,
then return
step 4
; or else, let
RB
=
RB
i
and end the algorithm.
5
Simulation Examples
In this section, the examples of Box-Jenkins gas furnace [11] and low voltage
electrical application [6] are applied to verify the effectiveness of EOCA-IFIM.
Case I. Box-Jenkins example
The Box-Jenkins gas furnace system is a SISO dynamic nonlinear process
with 296 samples. At each sampling time
k
, the input
x
(
k
) is the gas flow rate,
and the output
y
(
k
)isthe
CO
2
concentration. For verifying the robustness of
model, the sample set
Z
is added on a white gauss noise with 5dB signal-noise
ratio. Then the input is
X
=
x
(
k
−
1)
,x
(
k
−
2)
,y
(
k
−
1)
,y
(
k
−
2), and the output
is
y
(
k
).
Fig. 2.
Distribution of clusters under the input
y
(
k −
1)
,x
(
k −
1)
By EOCA, the minimum enhanced consistency index are
ita
AB
=0
.
8248 and
c
AB
= 9. Similarily,
ita
CD
=0
.
0211 and
c
CD
= 2. Thus the initial number
of clusters
c
0
is 2. The results of clustering by EOCA and FCM are shown
in Fig.2.
Search WWH ::
Custom Search