Physics-Informed Machine Learning of Chaotic Kuramoto Sivashinsky Systems

Abstract

Machine Learning has quickly become a powerful tool in the study of complex dynamical systems. Several papers have demonstrated how physical understanding can be included in neural networks to more accurately model complex physical systems. Here, physics-informed neural networks are used to model systems governed by the Kuramoto Sivashinsky equation. The Kuramoto Sivashinsky equation comes in different forms, and experiments are constructed to develop an intuitive understanding of the equations regimes. Additional experiments provide understanding of the performance of PINNs in Kuramoto Sivashinsky systems. Finally, improvements are suggested and tested.