Stable Linear System Identification with Prior Knowledge by Elastic Riemannian Sequential Quadratic Optimization

We consider an identification method for a linear continuous time-invariant autonomous system from noisy state observations. In particular, we focus on the identification to satisfy the asymptotic stability of the system with some prior knowledge. To this end, we first propose novel modeling in the form of a Riemannian nonlinear optimization (RNLO) problem. We ensure the stability by using a Riemannian manifold and additionally consider nonlinear constraints for the identified system to meet the prior knowledge. To solve the problem, we propose an elastic Riemannian sequential quadratic optimization (eRSQO) method. eRSQO is an improvement of RSQO proposed by Obara, Okuno, and Takeda (2020) with respect to the feasibility of subproblems to obtain a search direction; if the algorithm detects the infeasibility, eRSQO solves an alternative subproblem, which always has a feasible point, and computes a search direction. We prove the global convergence property of eRSQO under weaker assumptions than RSQO. Finally, we demonstrate the effectiveness of the proposed RNLO modeling and eRSQO method through numerical experiments.
15 pages, 4 figures