Chapter 11 . Input Selection for an Aigal Bloom Model
227
Table 11.2. Details of the Models Developed Using Each Combination of Input
Determination Technique
Unsupervis Supervised Model In uts Selected
ed method Method
No.
p
apriori
identificati
on
PCA
GA-ANN
Stepwise 2
GA-ANN 3
Stepwise 4
Anabaena lags I, 3, 4
Turb. lags 5, 6
Col.lags I, 2, 5, 7, 9
Temp. lags 2, 3, 4
Flow lags I, 4, 6, 8, 9
pH lags 1,2
Total P. lags 2, 3, 5, 7
Sol. P. lags 1,2,3,4,5,7,8,9
TKN lags 2, 3, 5, 6, 7, 8
Flow lags I, ... , 10
Turb. lags I, ... , 10
Temp.PC 1
Col. PCs 1,2
pHPC 1
Silica PCs 1,3,4
TKN PCs 1,5,6
Total P. PC 1
Sol. P. PCs 2, 3, 4, 5, 6
River Lev. PCs I, 2
Anabaena PCs 1,3,4,6
Flow PCs 1,2,3
Total P. PCs I, ... ,5
Temp. PCs 1,2
Stepwise SOM GA-ANN
Architecture Hidden Output
(~-~NY)
38 - 57 - 1
20 - 35 - I
22 - 36 - 1
10 - 52 - 1
5
layer
layer
nodes
node
28
Logistic Tanh
22 Tanh
7 Linear
15
Logistic Logistic
11 Tanh
9 Linear
16
Logistic Tanh
ITanh
19 Linear
6Tanh
.
46 Linear Lmear
20
Logistic
20 Tanh Logistic
20 Linear
Anabaen
a lags I,
5,7,12,
13,20,
23
Flow
lags 7,8,
10,13, 20 -6115,17,
1
20,22,
25
Turb.lag
1
Col.lags
1, 16
Silica lag
1