Data-driven system identification of unknown systems utilising sparse identification of nonlinear dynamics (SINDy)

Abstract This paper focuses on the application of experimental data-based system identification of unknown systems utilising sparse identification of nonlinear dynamics (SINDy). SINDy is used to detect the system dynamics in the three well-known nonlinear systems. Analyzed are SINDy’s abilities to accurately represent transient/steady-state behaviour, noise effect, and model structure. A sparse set of basis functions can effectively capture the dynamics of a system, according to the data-driven approach known as SINDy. The coefficients of these basis functions are determined via methods of sparse regression, and the final model is made up of a number of sparse ordinary differential equations. The findings demonstrate that SINDy, with sufficient time-series data, can capture both transient and steady-state phenomena. According to the analysis of the noise effect, SINDy’s performance declines as the system’s noise level rises. The feature library must contain the appropriate model structure in order for SINDy to function effectively. SINDy has the potential to extract unknown system dynamics from experimental data.