Information Technology Reference
In-Depth Information
In EOCA, Dipole Partition (DP) reduces the effect from noisy clusters with low
similarity. Then through the Relative Dissimilarity Measure-based Hierarchical
Clustering (RDM-HC), the effect of the unmerged clusters is considered [9]. Fi-
nally, in the Discriminating via Enhanced Consistent Criterion (DECC) process,
the effect of nearest neighbor in OCA is reduced. DP, RDM and ECC criteria
are referred to [4].
Then with
c
and
V
0
from EOCA, execute the FCM clustering, the clustering
centers and the fuzzy partition matrix
V
and the fuzzy partition matrix
U
are
gained. Followingly, by LSE, the muti-dimension fuzzy sets of clustering result
are fitted into the triangle membership functions. Thus the initial parameters of
Mamdani rules are formed.
3 Semantic Parameters Optimizing via (1+1) ES
In (1+1) Evolution Strategy (ES) [10], each rule is expressed as the gene and
denoted as
C
=(
a
1
,b
1
,c
1
)
...
(
a
p
,b
p
,c
p
)(
a
p
+1
,b
p
+1
,c
p
+1
), where
a
i
,b
i
,c
i
denotes
the left boundary point, the center and the right margin, respectively,
i
=
1
,
2
,...,p
+ 1 is the dimension of samples. Additionally, the fitness function con-
sists of two parts: the Covering Criterion (CC) and the Genetic Niching Principle
(GNP).
CC judge the satisfying degree of samples and includes three criteria: Average
Activation (AA), Average Covering Ratio of Satisfying Samples (ACRSS) and
Covering Ratio of Dissatisfying Samples (CRDS).
[Definition]
Satisfying (Dissatisfying) samples
Given the rule
R
i
,
z
k
is called as the satisfying sample iff it is satisfies:
A
i
(
X
k
)=
∗
(
A
i
(
x
1
)
,...,A
i
(
x
p
))
>
0
(1)
B
i
(
y
k
)
>
0
where
A
i
is the membership of premises;
is the fuzzy inference operator and
set product or minimum;
B
i
is the fuzzy sets of consequences;
X
k
=(
x
1
,...,x
p
)
is the input of
z
k
;
y
k
is the output of
z
k
.
In the same way,
z
k
is called the dissatisfying sample if
B
i
(
y
k
) = 0 is satisfied
in the formula (4).
Average Activation (AA)
The Average Activation (AA) is defined as follows:
∗
l
=1
N
R
i
(
z
l
)
Ψ
Z
N
(
R
i
)=
(2)
N
where
R
i
(
z
l
) is the activating degree of
z
l
by
R
i
.
Average Covering Ratio of Satisfying Samples (ACRSS)
The Average Covering Ratio of Satisfying Samples (ACRSS) is defined as
follows:
G
ω
(
R
i
)=
z
l
∈Z
ω
(
R
i
)
R
i
(
z
l
)
n
ω
(
R
i
)
(3)
Search WWH ::
Custom Search