Application of RBFNNs incorporating MIMO processes for simultaneous river flow forecasting

Abstract

Simultaneous flow forecasting using multi-input multi-output (MIMO) processes is an efficient technique for accurate flow forecasting on river systems. The present study demonstrates the capability of radial basis function neural networks (RBFNN) incorporating MIMO processes in simultaneous river flow forecasting. The river system considered in the present study was the Barak river system, Assam, India. Hourly concurrent discharge data were collected from the Central Water Commission, Shillong, India from multiple sections of the Barak river system. The forecasts were tested for short-range time horizons, i.e. 1, 3, 6 and 12 hours in advance, and a comparative analysis was done using the popular Nonlinear Autoregressive with Exogenous Inputs (NARX) time series model. The result shows that MIMO-NARX provided higher prediction accuracy than MIMO-RBFNN, even at longer lead times when compared to following various statistical criterions.