PERFORMANCE OF A VIRTUAL RUNOFF HYDROGRAPH SYSTEM

A Virtual Runoff Hydrograph System (VROHS) based on artificial neural network technology was designed and developed to generate runoff hydro graph. Data from 45 lab experiment sets were used to develop the VROHS . A recurrent back-propagation neural network was trained to generate runoff hydrograph. Twenty-nine of the 45 lab experiment sets were rand omly selected to train the network, while 16 experiment sets were sele cted to test the VROHS. It was found that the VROHS could predict the runoff hydrograph system very accurately for sets of input data (exper imental conditions) that it had never seen before. The values of the c orrelation coefficients and coefficient of determination for the testi ng sets ranged from 0.96 to 0.99 and 0.92 to 0.99, respectively. These high coefficient values demonstrated the good correlation between the observed data and the predicted data and also the high performance of the VROHS.