Determining Number of Input Nodes of Recurrent Neural Networks for River Flow Forecasting

In recent years, artificial neural networks have been applied in many fields. Reportedly, the structure of networks seriously affects the performance of the network model. A scheme for river flow forecasting with a recurrent neural network has been proposed in this paper which consists of the phase that determines of input patterns. Autocorrelation analysis is used to identify the number of input nodes of time series for training. The calculated number of neurons for an input layer is then used to construct the fully recurrent neural network forecaster. Computer simulations are presented to show the effectiveness of the scheme. The results obtained show that autocorrelation and cross correlation  analysis can be used to determine the number of input nodes. In terms of the performance statistics, an online training algorithm for fully recurrent neural networks has high forecasting accuracy.

 Keywords: Autocorrelation analysis, crosscorrelation analysis, network architecture, fully recurrent neural networks, number of input nodes, efficiency index, computational time.

 Corresponding author: E-mail: suwarin@rmut.ac.th, suwat@rmut.ac.th