Data-Driven Distributed Observer for Interconnected Systems

Abstract

This letter addresses the distributed observer design problem for linear interconnected systems with unknown dynamics using offline data. In contrast to classical model-based methods, the proposed approach constructs a distributed observer through a direct data-driven formulation without explicitly identifying a parametric state-space model. A key idea is to explicitly incorporate neighboring outputs so that the coupling terms can be canceled in the local error dynamics, rather than being treated solely as unknown inputs. Necessary and sufficient data-based conditions are derived for observer feasibility and asymptotic convergence. Furthermore, these direct data-driven conditions are shown to be equivalent to their model-based counterparts, thereby clarifying the relation between data-driven synthesis and model-based observer design. A multi-area power benchmark is employed to show the effectiveness of the proposed method.