Sparse Nonlinear Dynamics Models with SINDy, Part 2: Training Data & Disambiguating Models

Abstract

This video discusses data requirements for the Sparse Identification of Nonlinear Dynamics (SINDy) algorithm. Specifically, we discuss the required sampling rate and duration for clean data and how to compute SINDy for noisy data. The integral SINDy extension is described, which enables the discovery of a hierarchy of fluid and plasma models that are more or less complex than the standard Navier-Stokes and MHD models. Finally, we discuss the condition number of the SINDy regression and how this may be improved to better disambiguate between multiple consistent models.