| Hi-index | 0.00 |
We propose a simple method for the reconstruction ofslow dynamics perturbations from non-stationary timeseries records. The method traces the evolution of theperturbing signal by simultaneously learning the intrinsicstationary dynamics and the time dependency of thechanging parameter. For this purpose, an extra input unitis added to a feedforward artificial neural network and asuitable error function minimized in the training process.Testing of our algorithm on synthetic data shows itsefficacy and allows extracting general criteria forapplications on real-world problems. Finally, apreliminary study of the well-known sunspot time seriesrecovers particular features of this series, includingrecently reported changes in solar activity during lastcentury.