Modeling Inflows into System Reservoirs using Artificial Neural Networks

R. Chibanga, J. Berlamont, and J. Vandewalle (Belgium)


inflows; reservoir-routing; forecasting; feedforward; modeling; tributary-direct runoff


The paper set out to model and to predict inflows into a system of reservoirs for a study sub catchment in Zambia using artificial neural networks (ANNs). Working with data from the said sub catchment, several feedforward backprogation-artificial neural networks (FFBP-ANNs) are trained to learn the derived tributary-direct runoff, TrRO(t) in one instance and the Kafue River main flow, Q(t) series measured at the Kafue Hook Bridge (KHB) in another. To evaluate the forecasting performance of the selected ANNs comparison is made with the best Autoregressive Moving Average models (with exogenous inputs) ARMA(X). In both cases the ANNs give more robust forecasts over long term than the ARMA(X) models, thereby making ANNs a viable approach to reliably forecast inflows for long-term reservoir operations planning.

Important Links:

Go Back