Multivariate reservoir inflow forecasting using temporal neural networks

An experiment on predicting multivariate water resource time series, specif ically the prediction of hydropower reservoir inflow using temporal neural networks, is presented. This paper focuses on dynamic neural networks to ad dress the temporal relationships of the hydrological series. Three types of temporal neural network architectures with different inherent representati ons of temporal information are investigated. An input delayed neural netwo rk (IDNN) and a recurrent neural network (RNN) with and without input time delays are proposed for multivariate reservoir inflow forecasting. The fore cast results indicate that, overall, the RNN obtained the best performance. The results also suggest that the use of input time delays significantly i mproves the conventional multilayer perceptron (MLP) network but does not p rovide any improvement in the RNN model. However, the RNN with input time d elays remains slightly more effective for multivariate reservoir inflow pre diction than the IDNN model. Moreover, it is found that the conventional ML P network widely used in hydrological applications is less effective at mul tivariate reservoir inflow forecasting than the proposed models. Furthermor e, the experiment shows that employing only time-delayed recurrences can be the more effective and less costly method for multivariate water resources time series prediction.