Artificial Neural Networks Based Modeling of Stream Flows For Sindh Basin In Madhya Pradesh

Abstract

An artificial neural network with a feed forward back propagation algorithm has been employed for development of the model in the Sindh basin up to Madhikheda dam site. The daily rainfall and stream flow data for a period of 9 years from 1992 to 2001 has been used for training and testing the model. It has been found that a feed forward neural network with a single hidden layer and five neurons in the hidden layer is sufficient to model the stream flows at the dam site. The efficiency of the model varies between 70.36 % and 94.57 % with an overall efficiency of 85.14 % during calibration and between 63.02 % and 92.76 % with an overall efficiency of 84.42 % during validation. The percentage difference between the observed and computed peak flow during the calibration and validation period varies between 1.335 % and 0.319 % respectively. The correlation coefficient between the observed and computed flow series varies between 0.79 and 0.97. The ability of the three layer artificial neural network to relate the rainfall to the runoff response of the Sindh river basin up to Madhikheda dam site is demonstrated.