Seasonal Streamflow Series Forecasting Using Recurrent Neural Networks
Jônatas Trabuco Belotti, Lilian do Nascimento Araujo Lazzarin, Fábio Luiz Usberti, Hugo Valadares Siqueira · 2018
The present study compares the performance of different architectures of recurrent neural networks in the prediction of monthly seasonal streamflow series related to Brazilian hydroelectric plants. The architectures utilized are the proposals from Jordan and Elman and the Echo State Network. In the last case, the reservoir designs from Jaeger and Ozturk et al. were addressed, as well as the verification of the benefits in use the coefficient of regularization to calculate the weights of the output layer. The computational results showed that the ESN reached the best performance, mainly those with the regularization.