A New Neural Network Structure For Modeling Non-linear Dynamical Systems

摘要

In this paper a new two-layer linear-in-the-parameters feeforward network is presented termed the Functionally Expanded Neural Network (FENN). Its output error surface is shown to be uni-modal allowing high speed single run learning. It employs a least squares based learning algorithm for updating its output layer weights thereby linear basis functions emulating other universal approximators, namely the sigmodial, Gaussian, and polynomial-subset basis functions, have been proposed for inclusion in the network's single hidden layer. Both simulated chaotic (Mackey Glass time series) and real-world noisy, highly non-stationary (sunspot) time series data have been used to illustrate the superior modeling and prediction performance of the FENN compared with other recently reported, more complex feedforward and recurrent neural network based predictor models.