Data-driven Method for Identifying the Expression of the Lyapunov Exponent from Random Data
Xi Chen,Xiaoling Jin,Zhilong Huang
DOI: https://doi.org/10.1016/j.ijnonlinmec.2022.104268
IF: 3.336
2022-01-01
International Journal of Non-Linear Mechanics
Abstract:A data-driven methodology is proposed to identify the expression of the largest Lyapunov exponent to determine the stochastic stability of the system. The method is carried out by two successive processes: one is to estimate the largest Lyapunov exponent from random discrete data, and the other is to identify the simplest expression of the largest Lyapunov exponent which includes system and excitation parameters. The first process is based on a new definition of the Lyapunov exponent of the disturbed states by using the ensemble average in a finite time. The data of the disturbed states are obtained from the differences between the random data of a few samples of system states. The second process resorts to the dimensional analysis method and the sparse optimization algorithm. Five typical examples, including linear stochastic dynamical system, Duffing oscillator, viscoelastic system, nonlinear system with two degrees of freedom subjected to Gaussian white noise, and Duffing oscillator subjected to uniformly distributed excitation, are given to illustrate the effectiveness of the proposed data-driven method and the extensionality of the identified expression of the largest Lyapunov exponent including system and excitation parameters. Moreover, it is concluded that the proposed data-driven method needs a smaller amount of data by comparing the traditional method for estimating the largest Lyapunov exponent from random data.