Paper

The Efficiency of Input Determination Techniques in ANN for Flood Forecasting, Mun Basin, Thailand


Authors:
Tawee Chaipimonplin; Thaveesak Vangpaisal
Abstract
The selection of input variables from different techniques can provide different artificial neural network (ANN) model performances. This study utilizes an ANN model to forecast the water level at the M.7 gauge station for t+48 hour. Three objectives to be investigated are: (1) to compare the efficiency of four input determination techniques (cross correlation, stepwise regression, cross correlation with stepwise regression, and genetic algorithm); (2) to investigate the number of hidden nodes from 1-2n+1 node; and (3) to compare two different learning algorithms (Levenberg Marquardt-LM and Baysian Regularization-BR). Results demonstrate that the cross correlation and the cross correlation with stepwise regression techniques are best for selecting input variables to forecast water levels at t+48 hours at the M.7 gauge station. Additionally, the use of only one hidden node is sufficient for the ANN model, and LM and BR learning models perform similarly.
Keywords
Input Determination; Mun Basin; Flood Forecasting; Artificial Neural Network
StartPage
131
EndPage
137
Doi
10.5963/JWRHE0402002
Download | Back to Issue| Archive