Reduced-Order Model Accounting for High-Frequency Effects in Power Electronic Components

Abstract

This paper proposes a reduced-order model of power electronic components based on the proper orthogonal decomposition. Starting from a full-wave finite-element model and several snapshots/frequencies, the reduced-order (RO) model is constructed. Local field values (e.g. magnetic flux density, electric current density, magnetic or electric field) and global quantities (e.g. characteristic complex impedance, joule losses) can be determined for the intermediate frequencies with a very low computational cost and high accuracy. Particular attention is paid to the choice of the most suitable snapshots by means of three different greedy algorithms, the performance of which is compared. We adopt an automatic greedy algorithm that only depends on the RO model.

Fast Frequency and Material Properties Sweeps for Quasi-Static Problems

“…Moreover, to be more efficient over a wide number and range of frequencies, this technique should be coupled together with other standard model order reduction strategies as, for example, [14] and [15].…”

Proper orthogonal decomposition versus Krylov subspace methods in reduced-order energy-converter models

“…The set of the vectors H j constructs Krylov subspaces as H 0 = ↵, H 1 = ↵ 1 and so on. The Arnoldi's algorithm is used for generating projection basis from the Krylov subspace [13]. Hence, is built from the orthogonal basis of K n (↵, ).…”

“…Contrary to the time domain (3), the frequency domain solutions (4) are independent for different frequencies allowing a greedy approach to efficiently build the RB by computing the specific requested responses. In [13], the authors proposed an original greedy selection method to find the minimum number of snapshots M within a prescribed tolerance τ for the error on a quantity of interest. It solves the original system for well chosen frequencies, increases the reduced basis, and stops when the observed quantity has converged (e.g.…”

“…Indeed in the frequency domain, a uniform snapshot/frequency distribution is used and in the time domain, the bases are generated from a given number of time steps, fixed a priori and increased till desired accuracy is achieved in a trial and error approach. In this paper, we extend our frequency-domain algorithm [13] to the time domain and propose an original and efficient greedy algorithm in time domain. Furthermore, we compare both methods in time with projector operators generated by greedy approaches either in the frequency or time domain.…”

ParaExp Using Leapfrog as Integrator for High‐Frequency Electromagnetic Simulations

“…This choice is motivated by the linearity of the underlying governing equations. However, the solution of problems in frequency domain may require the resolution of very large linear systems of equations and this becomes particularly inconvenient for broadband simulations such that approximations like model order reduction are typically used (e.g., Floch et al, ; Paquay et al, ; Slone et al, ). The coupling with nonlinear time‐dependent systems and the computation of transients are other cases where time domain simulations outperform frequency domain simulations.…”