In this article, we propose two new iterative algorithms to solve the frequency-limited Riemannian optimization model order reduction problems of linear and bilinear systems. Different from the existing Riemannian optimization methods, we design a new Riemannian conjugate gradient scheme based on the Riemannian geometry notions on a product manifold, and then generate a new search direction. Theoretical analysis shows that the resulting search direction is always descent with depending neither on the line search used, nor on the convexity of the cost function. The proposed algorithms are also suitable for generating reduced systems over a frequency interval in bandpass form. Finally, two numerical examples are simulated to demonstrate the efficiency of our algorithms.
2 January 2023 in Linear Algebra and its Applications
This paper explores model order reduction (MOR) methods for discrete time-delay systems in the time domain and the frequency domain. First, the discrete time-delay system is expanded under discrete Laguerre polynomials, and the discrete Laguerre coefficients of the system are obtained by a matrix equation. After that, the reduced order system is obtained by using the projection matrix based on these coefficients. Theoretical analysis shows that the resulting reduced order system can preserve a certain number of discrete Laguerre coefficients of output variables in the time domain. We also derive the error bound in the time domain. Furthermore, in order to obtain the moments of the system, we approximate the transfer function of the discrete time-delay system by Taylor expansion. The basis matrices of the higher order Krylov subspace are deduced by a iterative process. Further, we prove that the reduced order system can match a desired number of moments of the original system. The error estimation is obtained in the frequency domain. Finally, two illustrative examples are given to verify the effectiveness of the proposed methods.
26 August 2019 in Japan Journal of Industrial and Applied Mathematics
In this article, we propose a new frequency-limited Riemannian geometric-nonlinear conjugate gradient model order reduction (MOR) method with the modified Armijo step-size control to solve the frequency-limited $\mathcal {H}_{2}$ optimal MOR problem of linear port-Hamiltonian systems. This problem is formulated as a Riemannian optimization problem on the product of several manifolds. Based on the geometric properties of the product manifold, the Riemannian gradient of the cost function is derived. Further, by scaling the vector transport of the product manifold, we design a new Riemannian spectral conjugate gradient direction, which is always descent for the cost function and is independent of the line search used and the convexity of the cost function. Meanwhile, a new modified Armijo step-size control strategy is presented. The resulting reduced system can preserve the port-Hamiltonian structure and the passivity of the original system. Two numerical examples are simulated to demonstrate the efficiency of the proposed method.
22 February 2022 in International Journal of Systems Science
This paper investigates model order reduction (MOR) methods of discrete-time systems via discrete orthogonal polynomials in the time domain. First, this system is expanded under discrete orthogonal polynomials, and the expansion coefficients are computed from a linear equation. Then the reduced-order system is produced by using the orthogonal projection matrix that is defined in terms of the expansion coefficients, which can match the first several expansion coefficients of the original output, and preserve the asymptotic stability and bounded-input/bounded-output (BIBO) stability. We also study the output error between the original system and its reduced-order system. Besides, the MOR method using discrete Walsh functions is proposed for discrete-time systems with inhomogeneous initial conditions. Finally, three numerical examples are given to illustrate the feasibility of the proposed methods.