In this work model order reduction (MOR) of differential algebraic equations (DAEs), with focus on structured, linear time-invariant systems, is investigated. As usual in the DAE-setting it is assumed, that the spectral projectors, which describe the structure of the model, are available. Those allow a separation of the actual dynamics and the involved algebraic equations which describe a constraint manifold to which the dynamics are bounded.
While H2 pseudo-optimal reduction by the Krylov-based pseudo-optimal rational Krylov (PORK) algorithm is applied to the strictly proper part of the transfer function, Lyapunov balanced truncation (BT) is used to find a minimal realization of the improper contribution. For this purpose the PORK algorithm, originally developed for the reduction of ordinary differential equations, is revisited in the context of strictly proper DAEs. As the original proof has to be modified, a detailed derivation of the PORK algorithm is presented. Furthermore the combination with adaptive MOR schemes like the stability-preserving, adaptive rational Krylov (SPARK) algorithm and the cumulative reduction (CURE) framework is discussed.
One of the main tools used in this thesis are generalized Sylvester equations. It is shown that they can be used to describe tangential-input rational Krylov subspaces, even in the case of a singular descriptor matrix. Moreover a formulation of the H2 inner-product of the transfer functions of two strictly proper DAEs via projected generalized Sylvester equations is presented. Those results are essential for the proof of H2 pseudo-optimality in PORK, but may also be useful in different contexts.
Finally an efficient overall-algorithm for the reduction of structured linear DAE-systems of arbitrary index is presented, which adaptively chooses appropriate interpolation data and reduced order. By means of several physically based models it is shown, that the proposed technique is applicable to common technical problems, regardless of properness or index of the given system. Furthermore the reduction of an artificially generated high-index system is demonstrated.
«
In this work model order reduction (MOR) of differential algebraic equations (DAEs), with focus on structured, linear time-invariant systems, is investigated. As usual in the DAE-setting it is assumed, that the spectral projectors, which describe the structure of the model, are available. Those allow a separation of the actual dynamics and the involved algebraic equations which describe a constraint manifold to which the dynamics are bounded.
While H2 pseudo-optimal reduction by the Krylov-ba...
»