| सारांश: | River debit prediction in many cases is much needed. Prediction system
is needed to see possibility dealing with the changes of nature condition in the
forthcoming time. In the water debit prediction, some analysis sign methods are
used. They are Artificial Neural Networks (ANN) and Fast Fourier Transform
(FFT).
Debit sign is transformed into frequency domain to find out the
frequency components. The result of transform gives clues about the existence of
periodicity cycles of raising and declining debit that is used for prediction.
Frequency filtering is employed to decompose the swelling periodicity on each
raising or declining debit. Predicted data with certain frequency are able to
determine the debit magnitude of the forthcoming time. River debit periodization
on each frequency component can be determine being long, middle, or short term
flood or dry period.
ANN that is trained for the sake of prediction using Levenberg Marquadt
algorithm has optimum architecture with 7 input neurons, 11 hidden layer neurons
and 1 out layer neuron. The result of ANN performance of long term debit
periodization is 2,048 days in dry period and 1.171 days in flood period. The
forecasted highest debit will take place on May 21st, 2011 � 7 days with the flood
debit of 2512 m3/second and the dry debit of 38,28 m3/second in August 2015.
Approaching some years before the days, with the additional data that found
afterward, it can be better predicted with middle term approach, and then for the
following months with short term.
|
|---|