Application of data mining techniques as a complement to natural inflow uni-variable stochastic forecasting - a case study : the Iguacu River Basin

This paper presents the results obtained from the utilization of a public dominion software that, through data mining and neural networks with Bayesian training is capable of laying the foundation for the selection of the most appropriate natural inflow forecast used in the PREVIVAZ stochastic modeling system. This technique utilizes precipitation information, forecasted and observed, a well as verified natural inflow data recorded over the weeks that precede the actual forecast target made at the water courses at the Foz do Areia and Jordao hydroelectric plants located in the Iguacu River Basin. The results obtained indicate that the usage of these tools can provide a simple and efficient solution to reduce natural inflow forecast errors on a weekly forecast basis for the Iguacu River Basin.