Application of Artificial Neural Networks for Short Term Rainfall Forecasting
Primena veštačkih neuronskih mreža za kratkoročno predviđanje padavina
Author
Mohanbhai, Kyada Pradip
Kumar, Pravendra
Keywords
artificial neural networkfeed forward algorithm
rainfall prediction
back propagation algorithm
veštačka neuronska mreža
algoritam direktne distribucije
predviđanje padavina
algoritam učenja sa povratnim širenjem
Metadata
Show full item recordAbstract
Accurate rainfall forecasting is very necessary for water resource
management. Recently, several modeling approaches have been investigated to perform
such forecasting task. In the present study, possibility of forecasting rainfall in Junagadh
has been analyzed through feed forward artificial neural network models. The 30 years
data has been used for training and testing the ANN networks. In formulating the ANN
based Predictive model, single and double hidden layers network have been constructed.
The performance of models have been evaluated using Correlation coefficient, Mean
Square Error, Normalized Mean Square Error, Akaike’s information criterion,
Coefficient of Efficiency and volumetric error, and two best suitable models (7-12-13-1
and 4-6-4-1) have been selected from case I and II for rainfall forecasting in Junagadh.
Based on the performance evaluation of the models, two models found suitable for
prediction of daily rainfall for the study area. Precizna prognoza padavina je neophodna za upravljanje vodenim
resursima. Skorije je ispitivano nekoliko pristupa modeliranju postupaka ovakvih
prognoza. U ovom istraživanju je analizirana mogućnost predviđanja padavina u
Junagadh kroz modele veštačkih neuronskih mreža (ANN) direktnih distribucija. Za
treniranje i testiranje mreža su korišćeni podaci za period od 30 godina. Pri formulisanju
prediktivnog modela, zasnovanog na ANN, konstruisane su mreže sa jednostruko i
dvostruko skrivenim slojevima. Performanse modela su ocenjivane upotrebom
koeficijenta korelacije, srednje kvadratne greške, normalizovane srednje kvadratne
greške, kriterijuma Akaike informacije, koeficijenta efikasnosti i zapreminske geške, a
dva najprilagođenija modela (7-12-13-1 i 4-6-4-1) su izdvojena iz slučajeva I i II za
prognozu kiše u Junagadh. Na osnovu ocene performansi ova dva modela su usvojena za
predviđanje dnevnih padavina u ispitivanoj oblasti.