Data-driven identification for nonlinear dynamic systems

For nonlinear dynamic systems, this paper investigates problems of identification and parameter estimation. These problems are critical in aerial, electromechanical, robotic and other systems. Analysis and control of physical systems imply the use of adequate mathematical descriptions, ensuring sufficient fidelity. Particular challenges occur if systems exhibit oscillations, limit cycles and instabilities. We apply multivariate polynomials and model-to-system mismatches measures to solve identification problems during dynamic governance. Physics-consistent nonlinear models are parameterised, truncated and validated using matrix factorisation schemes and algorithms. Heterogeneous measurements adverse the information content and obscure observed data. Singular value decomposition ensures algorithmic convergence and validity. Using simulations and experimental studies, a data-driven identification concept is demonstrated and validated.