Modeling chaotic systems: Dynamical equations vs machine learning approach

Chaotic systems are ubiquitous in the real world, but often analytical models remain inaccessible. We find that a machine-learning method known as reservoir computing'' provides an alternative feasible way for modeling chaotic systems rather than conventional dynamical equations. Specifically, we show that recurrence in temporal and spatial scales of the trained reservoir system are indistinguishable from that of an observed chaotic system. Furthermore, by sharing a common signal, dual synchronization between a chaotic system and its learned reservoir system can be achieved successfully. In the same manner, we show that the identical synchronization also emerges on their coupled system. These findings reveal that reservoir computing approach would be excellent candidate for modeling a great variety of chaotic systems. (c) 2022 Elsevier B.V. All rights reserved.

Automated summary

Chaotic systems are common in the real world, but it is often difficult to obtain analytical models. We propose a machine-learning method called reservoir computing as an alternative approach to model chaotic systems, which is more feasible compared to conventional dynamical equations.