Environmental Engineering Reference
In-Depth Information
and had a constant intensity and frequency. It is used to test the model sen-
sitivity and measurement errors. Due to the large number of combinations
(10 input and 10 output variables) a screening experimental design [19]
was used for testing the best performance developed ANN model (MLP
10-13-10, table 4.4). A 10-factor screening design, with all inputs as the
independent variables, was selected for the study [19]. For this number of
input variables number, 21 assays should be made [32]. h e coded values
for the inputs were -1 and +1, for negative and positive white noise incor-
porating, and 0 for unchanged experimental values. Table 4.8 shows the
i nal experimental design with twenty-one combinations, which is in diag-
onal form. h e i rst design contained coded values instead of real, experi-
mentally observed, results (base case), and every next two models present
the ef ect of one input parameter with applying noise in the range of ±SD.
h at is how each input parameter is tested to show the level of inl uence on
outputs. h e complete database (1390 data points) was used, for a total of
1390×21 = 29,190 cases.
h e inl uence of the input variables over the outputs is shown in table 4.9
by calculated SOS values for all twenty-one designs. Obtained SOS values
corresponded to the level of experimental errors, and also showed the sig-
nii cance of inputs' inl uence on outputs. Most of the output variables were
af ected by i ring temperature (which is expected, considering the variance
of this variable).
CSB
and
CSC
were found most af ected by SiO
2
, Al
2
O
3
and
CaO content. Signii cant impact of Na
2
O, MgO and K
2
O content can also
be observed, for both
CSB
and
CSC
. Besides i ring temperature, the most
Table 4.8
Custom experimental designs for sensitivity analysis.
Assay No.
SiO
2
Al
2
O
3
Fe
2
O
3
CaO
....
TiO
2
Te m p .
1
0
0
0
0
....
0
0
2-3
±1
0
0
0
....
0
0
4-5
0
±1
0
0
....
0
0
6-7
0
0
±1
0
....
0
0
...
...
...
...
....
....
...
...
18-19
0
0
0
0
....
0
0
19-20
0
0
0
0
....
±1
0
20-21
0
0
0
0
....
0
±1
Search WWH ::
Custom Search