Estimation in linear networks:: algebraic and graph-theoretic results

Abstract:

Recently, researchers have been increasingly concerned about inference problems in dynamical networks. This focus on inference is motivated by new challenges in managing infrastructures and other large-scale networks including network-security issues in physical and cyber-physical networks [1], structure-preserving modeling and dynamics-analysis of power networks, modeling and management needs in air traffic management, and inference of models in biology and economics. The recent studies on inference in dynamical networks suggest that estimators and their performance closely depend on the structure of the graph associated with the dynamics of these systems [2][3]. This observation has brought about an effort to relate optimal estimator performance for dynamical networks’ states/parameters to the network’s graph topology. However, the methods for this new approach to estimation characterization are still limited: early works have largely focused on initial-state estimation, or have assumed that only a single observation is used in estimation. Specifically, the network inference techniques have provided graph-theoretic conditions for system observability and have shown their application in the network security context [4][5]; they have also studied basic reproduction ratio estimability in a multigroup model of virus propagation. These conditions are associated with the estimator structure, and hence allow us design more secure network dynamics or an efficient control policy [6]. In section 2, we consider the estimation of non-random modes of a linear dynamical system from noisy, multi-observation impulse response data. We …

Año de publicación:

2012

Keywords:

Fuente:

google