Daily Reservoir Inflow Forecasting Using Time Delay Artificial Neural Network Models

Accurate real-time reservoir inflow forecasting is an important requirement for operation, scheduling and planning conjunctive use in any basin. In this study, Time Delay Artificial Neural Network (TDANN) models, which are time lagged feed-formatted networks with delayed memory processing elements at the input layer, are applied to forecast the daily inflow into a planned Reservoir (Almopeos River basin) in Northern Greece. The network topology is using multiple inputs, which include the one time lagged daily reservoir inflow values and the time lagged daily precipitation values from three meteorological stations which are inside the Almopeos river basin and a single output, which are the daily reservoir inflow values. The choice of the precipitation input variables introduced to the input layer was based on the cross-correlation. In the forecasting part of this study, predictions of one day ahead were investigated. The training of ANNs suitable for the current application is the cascade correlation algorithm. 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 results show a good performance of the TDANN approach and demonstrate its adequacy and potential for forecasting daily reservoir inflow. Key-Words: Reservoir inflow, Real-time forecasting, Time delay artificial neural networks