Sampled-data observer-based deterministic learning in noisy environments and its performance analysis

doi:10.1016/j.conengprac.2025.106720

Sampled-data observer-based deterministic learning in noisy environments and its performance analysis

2026-01-21

0

PRE

AI

DOI：10.1016/j.conengprac.2025.106720

Original

Original request for help

Save

Abstract

En 中文

This paper presents a learning approach for nonlinear dynamical systems using noisy sampled-output data. A sampled-data observer with a nonlinear gain structure is first designed to reconstruct the state trajectory, achieving both rapid convergence and reduced sensitivity to measurement noise. A deterministic learning process based on the reconstructed trajectory is then employed to identify the system dynamics. By exploiting the partial persistent excitation (PE) property of the radial basis function (RBF) neural network, a generalized exponential convergence model is derived for the perturbed linear time-varying (LTV) system associated with the identification process. This model explicitly relates the observer gain, network structure, and noise level to learning performance, providing theoretical guidance for parameter selection. Furthermore, the learned dynamics are reused to construct a non-high-gain observer, enabling accurate state estimation with low computational complexity in similar tasks. The proposed approach is validated through simulation and compressor aerodynamic instability warning experiments, demonstrating its capability for accurate learning and high-performance utilization of nonlinear dynamics under noisy conditions.

Keywords:

System identification

Deterministic learning

Dynamical system

Persistent excitation

Compressor instability

Sampled-data observer-based deterministic learning in noisy environments and its performance analysis

Abstract

Journal

Organization