Geology Reference
In-Depth Information
4.6.5 Neuro-Wavelet Model
In a study by Remesan et al. [
74
], a multi-layer feed-forward network type of
arti
cial neural network (ANN) and a DWT model were combined together to
obtain a NW model. The discrete wavelet transfer model is functioned through two
set of
filters, namely high-pass and low-pass
filters, which decompose the signal
into two set of series, namely detailed coef
cients (D) and approximation (A) sub-
time series, respectively. In the proposed NW model, these decomposed subseries
obtained from DWT on the original data directly are used as inputs to the ANN
model. The NW structure employed in the present study is shown in Fig.
4.18
. Here
we are describing the model details in terms of rainfall-runoff modeling. Antecedent
rainfall and runoff information of the study area in the Brue catchment were
decomposed into three series of detailed coef
cients (D) and three series of
approximation (A) sub-time series. Kisi [
49
] constructed a new series by adding
some relevant D values and constructed an effective series, using that along with an
approximation component as an input of ANN. We believe that this may lead to a
loss in information associated with individual data series, and that a better way is
splitting the original series in a low resolution level and using those data series
directly as inputs of the ANN. So, the present value of runoff has been estimated
using the three resolution levels of antecedent runoff and rainfall information (i.e.,
runoff and rainfall time series of 2-day mode (D
q
1
;
D
i
p
1), 4-day mode, (D
q
2
;
D
i
p
2),
8-day mode (D
q
3
A
i
p
1, where q denotes runoff,
p denotes rainfall, and i and j denote the number of antecedent data sets of rainfall
and runoff respectively.
D
i
p
3), and approximate mode A
q
1
;
;
Fig. 4.18 The proposed hybrid neural-wavelet scheme for rainfall-runoff modeling [
74
]
