Electerical Engineering, Southern Methodist University
In this paper a nonparametric neural network (NN) technique for prediction of future values of a signal based on its past history is presented. This approach bypasses modeling, identification, and parameter estimation phases that are required by conventional parametric techniques. A multi-layer feed forward NN is employed. It develops an internal model of the signal through a training operation involving past history of the considered signal. Training is performed using the back-propagation algorithm. The trained net is then used to do the forecast. Training is continued during operation to improve performance. The net performance is tested on signals generated by autoregressive (AR) and autoregressive moving average (ARMA) models of different orders and results are compared to optimal forecasts.