Distributed dynamic estimation for cyber-physical systems

In recent years, cyber-physical systems are gaining increasing attention due to their versatility of use in various application fields. Their peculiar characteristic of integration between computational aspects and physical processes determines the necessity of addressing new vulnerabilities and security issues. Attacks launched in the cyber domain can have consequences on the physical components, posing serious risks to industrial and daily life activities, such as transportation systems or healthcare. Cyber-physical systems are composed by a set of interconnected devices that take measurements from the physical world through sensors and aim at processing the measured data for some purpose. This thesis examines the problem of estimating the state of cyber-physical systems subjected to sensor attacks, known as secure state estimation, starting with a mathematical analysis of the problem based on the existing literature. Since the cyber-physical systems framework is intrinsically distributed, the measurement acquisition occurs in a decentralized fashion. On the other hand, the computational aspect of problem resolution can be handled in two different ways. Firstly, we focus on the centralized approach, such that the sensor measurements are sent to a single, central computing device, which processes the data with a global perspective. An observer algorithm is designed and examined for this purpose, and its performance is compared with a state-of-the-art algorithm. Subsequently, a decentralized strategy is employed also for computations, as it is considered more suitable for cyber-physical systems due to privacy and fault vulnerabilities arising from the presence of a unique computing device. An innovative observer tailored for solving secure state estimation in this setting is introduced and evaluated. Two example problems are addressed, a theoretical one and a potential real-world application, employing the aforementioned algorithms for numerical simulations. Considering that centralized observers yield superior performance compared to their distributed counterparts, albeit with various practical and security-related drawbacks, this work focuses on the development and evaluation of decentralized algorithms, for which a standard has not yet been established in the literature.