Geoscience Reference
In-Depth Information
In the RNN model, the training parameters such as learning rate,
momentum factor are progressively getting changed according to error
obtained by the network. The number of epochs is kept very large so that the
network will not terminate due to insucient number of epochs. Thus only
parameter to be changed is the number of recurrent neuron. The decrease
of SSE for RNN models is found higher in all runs when compared with
that of FFBPN models. It is found that the number of epochs is reducing
with decrease in number of neurons in hidden layer. In all the run, the
number of epochs required to train the RNN model are less compared to
FFBPN. In the operation model for the Vaigai basin, RNN models have
shown performance satisfactorily during validation phases. Also the data
set resulted in a better representation when an RNN model replaces the
ANN model using linear programming.
5. Conclusions
A study on intra neuronal, feedback network (RNN) proves to be a powerful
operation tool that is drawn on the most recent developments in artificial
intelligence research with a capability of not requiring assumptions about
the underlying population. In this paper, the neural network approach for
deriving operating policy for Vaigai reservoir is investigated. Overall results
showed that the use of neural network model in operation of reservoir sys-
tems is appropriate. This developed model can be implemented for future
operation of Vaigai system with the significance of no apriori optimiza-
tion for getting release from the dam needs to be tested. A time series
modeling on operation using LP with RNN provides a promising alterna-
tive and leads to better predictive performance than classical optimization
techniques such as linear programming. The training time for intraneu-
ronal network has been dramatically reduced comparing with that of BP
networks (interneuronal) discussed here. Further work can be extended to
three-dimensional information processing rather than from planar.
References
1. J. Sima, Back-propagation is not ecient,
Neural Networks
9
, 6 (1996)
1017-1023.
2. S. Jiri, Backpropagation is not e
cient,
JournalofNeuralNetworks
9
, 6 (1995)
1017-1023.
3. T. A. Clair and J. M. Ehrman, Variations in discharge and dissolved organic
carbon and nitrogen export from terrestrial basins with changes in climate: A
neural network approach,
Limnol. Oceanogr
41
, 5 (1996) 921-927.
