Chapter 10 . Aigal Species Succession in Rivers
203
10.4.2
Configuring the Neural Network Architecture and Predictability
Network training (Table 10.3) was done with various time-delayed input vectors.
The number of nodes in all time vectors was between 9 and 21 when the boundary
was given between 2 and 22, except for the case of a I-day-delay (case 10), which
satisfied criteria suggested from Hecht-Nielsen (1987). Using the 4-year data set,
there was little tendency for decreases of node number and mean squared error
(MSE) when the time-delay was increased. The MSE for network training
decreased as TDL increased by month with a fixed number of hidden layer nodes
(see Chon et al., 2000). However, among 8 time vectors, 4-day-delay inputs gave
a significant negative correlation between node number and MSE. Jeong et al.
(2001a) predicted time-series algal biomass in the lower Nakdong River using a
model with 3-day-delayed inputs; this was related to water residence time.
According to the ecological input data (i.e. number of exemplars, input
parameters, and their unseen relationships), an adequate TDL could be selected
variably.
Table 10.3. Nüde numbers and MSE für each time-delayed vector with 1,100
iterations , correlation coefficients between node number and MSE for each time
vector, and best-predicting network for 1994 algal dynamics.
C No-deJav
l-dav-dclav
2Sl:::!l5!!!l ':!!:!.I«!& ":2!!l:~ ,·Ibn'«..., 6- Nil.
MSE Nil. MSE Nu. MSE Nu. MSE
Nil.
MSE Nu. MSE Nu. MSE Nn. MSE
19
0.003
14
0.003
16
0.002
16
0.003
U
0,004
16
0.003
12
0.004
18
0.003
1
15
().OO~
15
0.004
B
0,003
18
0,002
19
(I.(IOJ
19
0.006
13
0.004
15
0,002
J7
0,004
2Il
0.l102
14
0.110.1
l5 0.0(14
18
O.lXlJ
21
O.l)()]
17
O,ooJ
J]
0.(1)2
20
0.002
18
0.002
14
11002
21
0,006
16
0.004
21
0,004
18
0.003
18 0.002
5
19
0.(103
17
0.006
21
0.002
15 o,oen
21
0,002
21
0,003
15
0,002
11
0,(103
Ii
13
0.003
IJ
0,004
19
0,001
11
0,004
16
0.004
21
0,006
lt 0,003 20 0,005
7
20
0,005
I~
0.004
I~
0.002
16
0,002
18
0,(0)
19
0,002
20
0.006
21
0,002
8
20
fHHJ2
Il!
O.c~.II'.
20
0.001
!? 0 .. 003
2(1.
o.t~)J
15 0.002
20 O.!JO)
19
O,(NJ~
9
12
0.007
20
0,00]
16
0,(0)
21
0,002
21
0.00]
21
0,003
la 0,003 16 0.003
JO
15
0.004
22
0.002
18
0,002
14 0,005
18
0.00]
21
O,{I02
10
0,004
13
0,003
U
20
0,003
22
0.0117
19
0.001
1:1 0,004
21
tl.OllJ
11
O,OOS
IS
0,004
18
0.003
12
15
0.004
15
O.(K14
15
0,002
11 0,001
14
tl.lIOJ
\2
0,004
11
0,006
15
0.003
IJ
20
0,003
18
0.00]
20
0,002
16
0,003
19
0,003
17 0,002
20 0.004
6
0.,005
14
11
0,(101
20
0.003
16
0.001
IS 41.0el4
n
0.006
2'1
0..006
17 41.002
2Il
0,00'
15
16
0.004
20
0,007
21
0,003
15 0,004
21
M)()J
9
0.006
21
0.002
20 0..002
Max.
20
0,001
22
0.(107
21
0.003
21
0.006
21
0.1106
21
o.J.106
21
0,006
IK 0,(103
Mil1.
11
0.002
13
0.002
13
0.001
14 0,002
11
0.001
9
0.,002
10
0.002
11
0.002
Mcan
0,004
0,004
0,002
0.003
0,003
tl.!Kl4
0.004
0,003
-0.58
0.01
-0,29
-0.15
-0.78
-0,24
0.,15
-0.39
(0.02" P
(p ~ 0,50)
tlJ,50 :~p
lp >0,50)
(0,001 "p)
(0.50." P
lp> 0,50)
(O,20>p
>0..111)
>0.20)
>4120!
>0.1O!