Geology Reference
In-Depth Information
(
)
X
N
y
¼
f
ð
x
Þ
¼
1
a
i
K
ð
x
i
;
x
Þ
b
ð
4
:
61
Þ
i
¼
where K is the kernel function,
a
i
and b are parameters, N is the number of training
data, x
i
are vectors used in training process, and x is the independent vector. The
parameters
owchart
on the SVM model selection in general rainfall-runoff modeling is shown in
Fig.
4.17
as per the study by Bray and Han [
14
].
As shown in Fig.
4.17
, the whole system is complex, and the selection of a better
model structure would require a large number of trial and error attempts, such as
making adjustments to rainfall scale factor,
a
i
and b are derived by maximizing their objective function. A
flow scale factor, cost factor, kernel
functions, and the parameters for the kernel functions. Many of these factors are
very sensitive and a wrong setting could result in an extremely long training time
and unstable results. Hence, it is very dif
cult to automate the selection process.
Bray and Han [
14
] have made an attempt to explore the relationships among
different factors which in
c
catchment. Before going for modeling, the modeler needs to make decisions about
the modeling scheme and how to pre-process the data, what kernel to use, and
several setting of parameters.
uence the SVM model development under a speci
Fig. 4.17 Schematic flow chart showing modeling procedures in SVM modeling (case study
rainfall-runoff modeling)