or resource management to enhance robust performance and improve the reliability index against time delays and packet dropouts, which are the inherent properties of open communication networks. Chapter 5 addresses the data‐driven degradation modeling of aging physical (actuators) and cyber (sensors) components of CPSs, and corresponding optimal maintenance plans to improve the reliability of CPSs. Chapters 6 and 7 investigate the cyber security of CPSs, introduce the general concept of cyberattacks, design vulnerability models, and risk assessment procedures, and develop game‐theoretic mitigation techniques and Bayesian‐based cyberteam deployment strategies.
More specifically, Chapter 1 summarizes the evolution from the traditional physical system to the CPS and provides an overview of dynamic and dependent behaviors to be addressed in the subsequent chapters of the book. The introduction discusses some important and recent challenges in improving traditional physical systems in terms of CPSs, popular research trends in evaluating the impacts of CPSs on society, and opportunities for enhancing the performance of realistic applications, which are primarily network control systems. The detailed properties, requirements, and vulnerabilities of utility systems are also introduced. The reasons why the proposed modeling techniques work is important in a field that would be difficult to deal with if the cyber and physical domains were treated separately.
In Chapter 2, readers acquire the basic knowledge to be used in data‐driven statistical modeling, the estimation of the probabilistic CPS state, and a comprehensive framework for conducting reliability analysis of CPSs. In addition, this chapter introduces how to use to historical data to validate the performance of the proposed CPS model, and how to use performance indexes to facilitate the resilient design of CPSs. Moreover, it also demonstrates a real‐time test platform for various industrial applications and the standard procedures for improving real‐time criteria.
Chapter 3 focuses on the stability of CPSs, where decision makers perform dynamic control and adaptation based on real‐time data from sensors. It provides two examples of the design of the controller parameters for robust system performance. The first example illustrates the development of adaptive control for wide‐area measurement power systems, where communication delays are predicted to provide delay compensation for additional frequency stability. In addition, the integration of control theory, power engineering, and statistical estimation is discussed. The second example is an extension of wide‐area measurement power systems from a dedicated communication network to open communication networks, where occurrences of communication delays and packet dropouts result in the failure of the power management system from renewable energy resources. Explicit and implicit methods are then designed for system integration, analysis, and improvement.
Chapter 4 illustrates a system‐of‐systems framework for the reliability of distributed CPS accounting for the impact of degraded communication networks. This is quite different from the focus of Chapter 3, which mainly covers the stability of CPSs from a control perspective. Based on the collected dataset, the degradation path of open communication networks is described in terms of stochastic continuous time transmission delays and packet dropouts. A distributed generation system with open communication infrastructure is used as an example, which is a multi‐area distributed system that is more complicated than the single‐area power system presented in Chapter 3. An optimal power flow model is proposed to generate consecutive time‐dependent optimal operation scenarios for a distributed CPS. Quantitative analysis is carried out to evaluate the effect of networked degradation on the reliability indexes of CPSs, e.g., energy not supplied and operation cost. A prediction method for reconstructing missing data is proposed to mitigate the influence of packet dropouts, which is universal and applicable to most current industrial applications.
Chapter 5 models the functional dependence between stochastic aging actuators and sensors within their operating environments. This dependence is considered in the time domain, causing a distinct degradation status in the actuators and sensors. Reliability modeling of the stochastic effects and effective maintenance activities are discussed for different types of CPSs, including the cooling system in a nuclear power plant, a one‐area energy system with a single generation group, and a multi‐area energy system with several different generation groups.
Chapter 6 explores the concepts, principles, practices, components, technologies, and tools behind risk management for cybersecurity of CPSs, providing practical experience through a realistic case study that focuses on the methodologies available to identify and assess such threats, evaluate their impact, and determine appropriate measures to prevent, mitigate, and recover from any threat or disruptive event so that the operations and profitability of the organizations are maintained and maximized.
Chapter 6 presents the framework of CPSs under cyberattacks from a game‐theoretic perspective, which makes use of statistical data to model the behavior of cyberattacks and study the dynamic game between the network defender and attacker at the system level. For current utility CPSs, cyber threats from supervisory control and data acquisition (SCADA) systems, and spear‐phishing attacks on the accounts of internal employees to gain access to dedicated communication networks are investigated in Chapters 6 and 7. In addition, these chapters focus on how to modify the basic modeling techniques presented in Chapter 2 to describe cyber vulnerabilities, such as seizing the SCADA system under control, disabling/destroying IT infrastructure components, and denial‐of‐service (DoS) attacks on the control center in smart grids. Based on historical data for IT security spending, the cost of launching distributed DoS attacks, and the occurrence probability of cyber event losses, the contest intensity between the attacker and defender can be accurately predicted for the next period to guide the design of an effective network protection plan, that is, a game‐theoretic protection plan and a Bayesian‐based cyberteam deployment.
Chapter 7 investigates sequential control problems (i.e., sequential cyberteam deployment) in modern CPSs by introducing an adversarial cost sequence with a variation constraint. Chapter 6 reviews the data‐driven vulnerability model, and Chapter 7 deals with the dataset of the arrival time of cyberattacks, which uncovers the statistical pattern of attackers. To solve such problems, a fundamental idea is to first obtain sampled parameters for the arrival model of cyberattacks from the posterior distribution of realistic cyberattack arrival records. The reinforcement learning model for estimating parameters is formulated as a partly parameterized Bayesian model. As a result, the sampled parameters are used instead of the true parameters. The paradigm of this framework can also be applied to other classical models, although specific models are used here for illustration purposes. Next, a Bayesian multi‐node bandit is built to cope with the problem, and an online learning algorithm (the Thompson‐Hedge algorithm) is forwarded to retain a converging regret function that is a function of the cyberteam deployment. By comparison with the existing algorithm, the convergence rate of the regret function in the proposed algorithm is found to be superior.
