DYNAMIC PREDICTIONS FROM TIME-SERIES DATA - AN ARTIFICIAL NEURAL-NETWORK APPROACH

A hybrid approach, incorporating concepts of nonlinear dynamics in art ificial neural networks (ANN), is proposed to model a time series gene rated by complex dynamic systems. We introduce well-known features use d in the study of dynamic systems time delay tau and embedding dimensi on d - for ANN modeling of time series. These features provide a theor etical basis for selecting the optimal size for the number of neurons in the input layer. The main outcome of the new approach for such prob lems is that to a large extent it defines the ANN architecture, models the time series and gives good prediction. As a consequence, we have an integrated and systematic data-driven scheme for modeling time seri es data. We illustrate our method by considering computer generated pe riodic and chaotic time series. The ANN model developed gave excellent quality of fit for the training and test sets as well as for iterativ e dynamic predictions for future values of the two time series. Furthe r, computer experiments were conducted by introducing Gaussian noise o f various degrees in the two time series, to simulate real world effec ts. We find that up to a limit introduction of noise leads to a smalle r network with good generalizing capability.