Study on replication of a nonlinear dynamical system’s trajectory using a machine learning technique

Abstract

Nonlinear system exhibits various solution orbits depending on varying parameters. It is important to detect the system’s behavior. In some cases, however, the mathematical modeling of the dynamic is completely unknown. By using a recent advance in the Machine Learning technique named Reservoir Computing, we replicate the solution orbits based only on data collected along with time evolution. We numerically confirm the effectiveness of Reservoir Computing in time series prediction.