Get documents about "Fuzzy Clustering Neural Network Based on Adaptive Dynamic Objective
Document Sample
34 11 2008 6
Vol.34 No.11 Computer Engineering June 2008
· · 1000 3428(2008)11 0035 03 A TP183
1,2 1,2 1 1
(1. 214122 2. 214405)
Fuzzy Clustering Neural Network Based on
Adaptive Dynamic Objective Function
BAO Fang1,2, PAN Yong-hui1,2, XU Wen-bo1, SUN Jun1
(1. School of Information Engineeing, Southern Yangtze University, Wuxi 214122; 2. Jiangyin Polytechnic College, Jiangyin 214405)
Abstract This paper proposes a novel adaptive dynamic objective function of fuzzy clustering algorithm, the new objective function integrates
the clustering characteristic of input space and the real time approximate characteristic of output space, so importing felicitous adaptive feedback
factors into the objective function. Extraordinary neural network to implement the fuzzy clustering algorithm is proposed. The new algorithm has
better performance in stable convergence rate, convergence speed, and threshold sensitivity compared with traditional fuzzy clustering algorithm.
Experiments show the algorithm provides more efficient and more stable application worthiness.
Key words fuzzy clustering; neural network; objective function; adaptive; dynamic; address selection strategy
1 ADFCNN)
2
X={x1,x2, ,xN} c ( )S1,S2, ,Sc
FCM [1]
µik(1 i C, 1 k N) xk Si
c- U={µik|1 i c, 1 k
n} Si(1 i c) vi
O xk Si
X={x1,x2, ,xN} V={vi,1 i c} ok vi dik=D(xk, vi)
C-
c-
c N m 2
(AO) (NN) J m = ∑ ∑ (uik ) xk − vi (1)
i =1 k =1
(EC)
c
∑ uik = 1,1 k N, 0 uik 1
i =1
[2]
m
[1,5]
? min{J m (U ,V )}
[3-4]
(60474030)
(1970 )
(Adaptive Dynamic Fuzzy Clustering Neural Network, 2007-09-02 E-mail baofang@mail.jypc.org
35
1
uik = ,1 k N ,1 i c (2)
c xk − vi
∑( )(2 /( m−1))
j =1 xk − v j
N
m
∑ uik xk
vi = k =1
N
,1 i c (3)
m
∑ uik (1) FCM
k =1
Vi0 zi (i=1,2, ,c)
Vi0
2.1 zi 1
(2) α w
[5] (3)
vi,α,w
(4) vi,w,α (8) uik
(5) (3) (4)
1 N
Ek = ∑ tk − ok
2
(9)
2 k =1
[6] 3
3
3.1
1 4 BP
1
u1k u2k uck
X={x1,x2, ,xN} T={t1,t2, , tN } y ki
yk2 y kc 2
O={o1,o2, , oN} y k1
αi
1 vi zi(i=1,2, ,c)
αi
d1k d2k dck 1
yi − zi = α i ( xk − vi ), k = 1, 2,
ˆ ,N (4) vi
2
c x k1 x k2 x kp
yk = ∑ α i ( xk − vi ) + zi , k = 1, 2,
ˆ ,N (5)
i =1
1
3 Wi i
W ij 1 i C, 1 j p p P P
C 1
2
yk = wi xk (6) Dik = xk − vi i
1
vi 2 C 1
c
4 yk = ∑ α i ( xk − vi ) + zi
ˆ
i =1
c N 2 2
J = ∑ ∑ uik ( xk − vi
m
+ yk − y k )
ˆ (7) 1 2 αi
i =1 k =1
FCM
U
(8)
vi m C
vi,
U J
α,w uik
1
uik = 1/( m−1)
,1 k N ,1 i c ( 8) 3.2
⎛ x −v 2 + y − y ˆk
2 ⎞
c
∑⎜
k i k ⎟ C 1
j =1 ⎜ x − v
2 2 ⎟
⎝ k j + yk − yk
ˆ
⎠ Wi i
2.2 W ij 1 i C, 1 j p
36
p i yki = wi xk 6 7 FCM
2
yki
0.7
0.6
wi
0.5
0.4
xk1 xk2 xkp 0.3
2 0.2
3.3 0.1
0.0
0 100 200 300 400 500 600 700 800 900 1 000 1 100 1 2001 3001 400 1 500
/
c
o = ∑ uik yki (10) 4 ADFCN N 1
i =1
(9)
2.5
3 2.0
1
yki 1.5
uik
xk C
1.0
0.5
3 0.0
0 100 200 300 400 500 600 700 800 900 1 000 1 100 1 2001 3001 400 1 500
/
4
5 ADFCN N 2
[7]
20 8
0.16
3 1 0.15
0.14
1
0.13
0.12
1 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00
2 0.8 0.87 0.89 0.82 0.79 0.80 0.75 0.35 0.79 0.11
3 0.69 0.93 0.22 0.75 1.00 0.80 0.49 0.66 0.73
4 0.92 0.80 0.89 0.92 0.89 0.80 1.00 1.00 0.81 0.10
5 0.87 0.93 1.00 1.00 1.00 1.00 1.00 1.00 0.96
6 0.80 0.72 0.89 0.82 0.89 0.80 0.75 1.00 0.83 0.09
7 0.67 0.72 0.67 0.66 0.67 0.60 0.49 0.66 0.69 0 100 200 300 400 500 600 700 800 900 1 000 1 1001 2001 3001 400 1 500
8 0.72 0.80 0.78 0.75 0.78 0.80 0.75 0.66 0.75 /
9 0.60 0.60 0.56 0.58 0.56 0.60 0.49 0.66 0.58
10 0.47 0.47 0.44 0.41 0.44 0.40 0.49 0.35 0.51 6 FCM 1
11 0.40 0.40 0.33 0.33 0.33 0.40 0.49 0.35 0.49
12 0.27 0.33 0.33 0.24 0.22 0.20 0.24 0.33 0.40 0.8
13 0.20 0.20 0.22 0.17 0.11 0.20 0.20 0.20 0.32
14 0.07 0.07 0.11 0.07 0.09 0.12 0.05 0.11 0.30 0.7
15 0.08 0.93 0.56 0.92 0.89 0.60 0.24 0.33 0.51
16 0.92 0.72 0.67 0.66 0.56 0.80 0.75 0.69 0.59 0.6
17 0.67 0.60 0.89 0.82 1.00 0.80 0.75 0.29 0.53
18 0.32 0.40 0.67 0.33 0.33 0.80 0.75 0.35 0.61 0.5
19 0.52 0.80 0.67 0.82 0.78 0.60 0.49 0.82 0.55
20 0.87 0.72 0.89 0.92 0.89 0.40 0.49 0.29 0.61 0.4
FCM 0.3
0.2
8
0.1
ADFCNN 0.0 0 100 200 300 400 500 600 700 800 900 1 000 1 100 1 200 1 3001 400 1 500
FCM 98.2%, 80.5% /
ADFCNN FCM 7 FCM 2
18.3% 40
4 5 ADFCNN
37
Related docs
