Deduzindo sistemas dinâmicos a partir de dados

Abstract

This study aims to use an algorithm capable of deducing the behavior of dynamic systems from data. The methods used in this algorithm were the Runge-Kutta-Fehlberg methods and the SINDy method, based on the collection of data on the system’s positions and velocities. The regression process used in the SINDy algorithm was the least squares sequencing method and the programming language used was Python. The main dynamic systems that were studied in this study were the Lorenz attractor and the damped pendulum. It has been concluded that the SINDy method is able to deal with different cases of the damped pendulum equations and their modifications, but it was not able to make predictions after a long time interval for the Lorenz attractor and a sparsification adjustment is needed for more reliable results in the case of damped oscillators.