A time delay artificial neural network approach for flow routing in a river system

Abstract. River flow routing provides basic information on a wide range of problems related to the design and operation of river systems. In this paper, three layer cascade correlation Time Delay Artificial Neural Network (TDANN) models have been developed to forecast the one day ahead daily flow at Ilarionas station on the Aliakmon river, in Northern Greece. The networks are time lagged feed-formatted with delayed memory processing elements at the input layer. The network topology is using multiple inputs, which include the time lagged daily flow values further up at Siatista station on the Aliakmon river and at Grevena station on the Venetikos river, which is a tributary to the Aliakmon river and a single output, which are the daily flow values at Ilarionas station. The choice of the input variables introduced to the input layer was based on the cross-correlation. The use of cross-correlation between the ith input series and the output provides a short cut to the problem of the delayed memory determination. Kalman's learning rule was used to modify the artificial neural network weights. The networks are designed by putting weights between neurons, by using the hyperbolic-tangent function for training. The number of nodes in the hidden layer was determined based on the maximum value of the correlation coefficient. The results show a good performance of the TDANN approach for forecasting the daily flow values, at Ilarionas station and demonstrate its adequacy and potential for river flow routing. The TDANN approach introduced in this study is sufficiently general and has great potential to be applicable to many hydrological and environmental applications.