Geoscience Reference
In-Depth Information
competing for scarce water supplies. Increasing basin water demands are
placing additional stresses on the limited water resources and threaten its
quality. Many hydrological models have been developed to problem of reser-
voir operation. System modeling based on conventional mathematical tools
is not well suited for dealing with nonlinearity of the system. By contrast,
the feed forward neural networks can be easily applied in non-linear opti-
mization problems.
Jiri
2
proposed the back-propagation learning algorithm for multilayered
neural networks, which is often successfully used in practice, appears time
consuming even for small network architectures or training tasks. Clair and
Ehrman
3
used a neural network approach to examine relationships between
climate and geography on discharge and Dissolved Organic Carbon (DOC)
and Dissolved Organic Nitrogen (DON) from 15 river basins in Canada's
Atlantic region over 10 years period. They emphasized the importance of
the evapotranspiration-precipitation link in establishing basin discharge,
because even large increase in precipitation can lead to decrease the dis-
charge when accompanied by higher temperature. Yang
et al
.
4
presented a
flood forecasting procedure by integrating linear transfer function (LTF),
autoregressive integrated moving average (ARIMA) model and ANN. They
illustrated the integrated method and a stand-alone ARIMA model applied
to Wu-Shi basin, Taiwan. The results obtained from these two models were
compared. It was concluded that the integrated ANN model, which con-
sists of ANN, LTF, and ARIMA model is appropriate for watershed flood
forecasting. Chandramouli and Raman
5
developed for optimal multireser-
voir operation, a dynamic programming-based neural network model. The
training of the network is done using a supervised learning approach with
the back-propagation algorithm. Rules derived for the three reservoirs using
dynamic programming-neural network model gave better performance than
dynamic programming-regression model. This paper gives a short review on
two methods of neural network learning and demonstrates their advantages
in real application to Vaigai reservoir system. The developed ANN model is
compared with LP model for its performance. This is also accomplished by
checking for equity of water released for irrigation purpose. It is concluded
that the coupling of optimization and heuristic model seems to be a logical
direction in reservoir operation modeling.
2. Study Area and Database
The model is applied to Vaigai basin (N 9
◦
15
-10
◦
20
and E 77
◦
10
-79
◦
15
)
which is located in south of Tamil Nadu in India (Fig. 1). The catchment
