Nonlinear set membership prediction of river flow

In the present paper an extension of a nonlinear set membership prediction method previously proposed by the authors is applied to a river flow prediction problem. This method does not require the choice of the functional form of the model used for prediction, but assumes a global bound on the gradient norm of the regression function defining the model. In the paper it is shown how to simply extend the method in order to use local bounds on the regression function gradient norm instead of a global ones. This local information may be relevant in improving prediction performances. The method is then used for the univariate prediction of the time series consisting of the mean daily discharges of the Dora Baltea river in northern Italy, taken from year 1941 to year 1979. The obtained prediction performances are compared with those obtained by means of the global nonlinear set membership method, of neural networks and of local linear approximation techniques previously used by other authors for this time series.