Journal of Control, Automation and Electrical Systems
https://doi.org/10.1007/s40313-020-00633-5
Observability of Network Systems: A Critical Review of Recent Results
Arthur N. Montanari
1
· Luis A. Aguirre
2
Received: 13 December 2019 / Revised: 28 July 2020 / Accepted: 4 August 2020
© Brazilian Society for Automatics–SBA 2020
Abstract
Observability is a property of a dynamical system that defines whether or not it is possible to reconstruct the trajectory temporal
evolution of the internal states of a system from a given set of outputs (measurements). In the context of network systems,
two important goals are: (i) to determine if a given set of sensor nodes is sufficient to render the network observable; and
(ii) what is the best set of sensor nodes among different available combinations that provide a more accurate state estimation
of the network state. Alongside Kalman’s classical definition of observability, a graph-theoretical approach to determine the
observability of a network system has gathered a lot of attention in the literature despite several following works showing that,
under certain circumstances, this kind of approach might underestimate, for practical purposes, the required number of sensor
nodes. In this work, we review with a critical mindset the literature of observability of dynamical systems, counterpoising
the pros and cons of different approaches in the context of network systems. Some future research directions for this field are
discussed and application examples in power grids and multi-agent systems are shown to illustrate our main conclusions.
Keywords Observability · Dynamical systems · Network systems · Sensor placement
1 Introduction
The mathematical modeling of dynamical systems is a fun-
damental framework in engineering that provides a means
to analyze aspects of a system, such as its stability, con-
trollability or observability, and thereafter design control
laws for practical applications (Chen 1999; Khalil 2002).
However, being designed for the most part with systems
of low-dimensional order in mind, classic control theory
methods are not efficient, or even feasible, for large-scale
systems, such as interconnected (networked) dynamical sys-
tems. This practical limitation has led control theory notions
to be adapted, optimized, or even redefined, in the literature
for high-dimensional applications (Chen 2014).
A specific, but recurrent, type of high-dimensional system
can be defined as networks. A network is a set of nodes inter-
B Arthur N. Montanari
montanariarthur@gmail.com
Luis A. Aguirre
aguirre@ufmg.br
1
Graduate Program in Electrical Engineering of the
Universidade Federal de Minas Gerais (UFMG), Av. Antônio
Carlos 6627, 31270-901 Belo Horizonte, Minas Gerais, Brazil
2
Depto. de Engenharia Eletrônica, UFMG, Belo Horizonte,
Brazil
connected by edges, in which information flows among its
elements through pairwise interactions. It can be mathemati-
cally modeled by graph structures, which allow a wide range
of useful metrics and algorithms of graph theory (Newman
2010; Chen et al. 2013; Bullo 2016). For instance, graph the-
ory can be used to assess the robustness to spreading failures
in power systems (Zhang et al. 2014; Schäfer et al. 2018) or
biological networks (Schimit and Monteiro 2009; Gilarranz
et al. 2017).
Up to the end of the twentieth century, it was believed
that real-world interconnected systems, such as neuronal,
social, communication, traffic, and energy networks, and
even the Internet, were composed of stochastic connec-
tions among its nodes. However, works over the last two
decades highlighted that most of real-world networks share
similar topological characteristics—not being purely ran-
dom, nor purely regular (Watts and Strogatz 1998; Barabási
1999, 2009). Complex networks, therefore, are a sub-
class of mathematical models derived from graph the-
ory, in which topological structures (graphs) show recur-
rent patterns that are found in the most diverse real net-
works present in nature and engineering (Chen et al.
2013; Barabási and Pósfasi 2016). Based on these find-
ings, the last years have been flooded with studies about
complex networks models, such as scale-free networks
123