Year: 2026
Type: preprint
Abstract: Ill-conditioned periodic transformations can hinder dense Floquet reconstruction for discrete-time linear time-periodic systems. We construct a surrogate using a fixed real monodromy root and sequential phase-wise least squares to fit periodic input and output maps from full-state and input data. Exact input–output equivalence is not imposed. Under the stated excitation, independence, and moment assumptions, the population input regression targets the original periodic input matrices. Completion requires finite predictions with averaged deviation-normalized error below one. Tests on a structured commuting ensemble with analytical roots, dimensions 2–8, and periods 3–501 show that the sur-rogate completes every trial. The dense canonical reference has lower conditional median prediction errors but completes only 1.3% of noiseless trials at n = 8, P = 501. For a shared surrogate-designed gain at n = 8, P = 251, boundary estimates within 30% of the numerically estimated true-system boundary are returned in 92.6% of surrogate trials versus 2.6% of reference trials. These results identify an accuracy–availability trade-off for the tested implementations.
Source: Research Square
Authors: Onurcan Yilmaz, Ismail Uyanik
Institution: Hacettepe University
Language: English
Subfield: Control and Systems Engineering
Field: Engineering
Domain: Physical Sciences
Open Access status: green
OpenAlex ID: W7225921365
