:   ..
:  
:  119
:  
:  2026
:   .. // . - 2026. - . 119. - .91-137.
:   , , , ,
(.):  model order reduction, observability gramian, subgramian, nonlinear approximation, Carleman linearization
:   . , (), . $H_2$- . , . , . , , . , , . , , .
(.):  The paper considers the use of the observability Gramian decomposition by the eigenvalues of the dynamics matrix of a linear dynamical system for solving the problem of model order reduction. The concept of double truncation is proposed, which assumes the simultaneous elimination of both the eigenvalues of the dynamics matrix (modes) and the state variables from the dynamical system. A theorem on the relationship between the quadratic Lyapunov function and the $H_2$-norm of the transfer function of a linear dynamical system of full and reduced dimension is formulated and proved. Based on this theorem, several methods have been proposed to assess the degree of significance of the modes and states being deleted, and an iterative algorithm for the double truncation method has been developed. The results obtained are extended to the case of reducing the dimension of autonomous nonlinear approximations of dynamical system models linearized by the Carleman method. The effectiveness of the proposed reduction method has been demonstrated during computational experiments conducted on three linear models and one quadratic approximation. In the linear case, the double truncation method provides a consistently good approximation of the original system, but is inferior to the methods of modal and balanced truncation in both accuracy and speed. In the case of autonomous nonlinear approximations, due to the implementation of the selective-by-states reduction, double truncation makes it possible to sufficient increase the accuracy of reduction in comparison with other known truncation methods.

PDF

: 23, : 7, : 4.


© 2007.