【作者】 刘珊；

【导师】 胥布工；

【作者基本信息】 华南理工大学 ， 控制理论与控制工程， 2019， 博士

【摘要】 信息物理系统(cyber physical system,CPS)是近年来提出的概念,它是指将计算、通信和物理过程紧密融合所形成的系统。这样的系统可以将感知、通信和计算技术嵌入到物理设备中,充分实现分布式传感、信息传输、智能信息处理、实时反馈和控制。因此,信息物理系统被应用在很多领域中,包括化工、电网、配水、汽车和智能家居等。信息传输是信息物理系统中最基本的功能,而有线或无线的传输设备会受到来自信息空间的各类攻击影响,从而引起信息传输的错误甚至失败,进而造成控制性能的下降或系统的不稳定运行。因此,信息物理系统的安全问题受到越来越多研究者的关注,相关的研究成果也从最初的通信领域扩展到控制领域。关于信息物理系统的安全问题主要分为两大类:信息安全和安全控制理论。前者主要关注加密和数据处理;而后者研究网络攻击如何影响控制系统的物理动态。信息安全的研究不能够完全阻止攻击者对控制系统造成影响,因此安全控制理论的研究也是有必要的。安全控制理论是基于攻击模型来展开研究的,如何建立合适的攻击模型很重要。在前人的研究中,最常见的攻击有两大类。第一类是拒绝服务(denial-of-service,DoS)攻击,另一类则是欺骗攻击,而虚假数据注入(false data injection,FDI)攻击是欺骗攻击的典型代表。最近,针对信息物理系统在这两类网络攻击下的安全控制研究已受到越来越多的关注,也取得了一定的进展,但仍存在许多不足之处。例如,现存在的方法对于拒绝服务攻击下的信息物理系统的弹性控制研究缺少考虑信息层状态转移概率未知的情况,且对攻击行为的研究也较单一,即多数研究只考虑单一攻击类型下的信息物理系统的控制问题。此外,缺乏考虑引入事件触发器以及系统存在非线性的影响等。本文针对这些不足之处,从控制系统的稳定运行、性能优化以及实际应用等方面展开研究,具体的研究内容如下:1.针对CPS在DoS攻击下的安全问题,提出了基于博弈的H_?控制策略算法。一方面,考虑控制性能的零和博弈,利用性能指标设计信息层的混合策略;另一方面,考虑攻防策略的零和随机博弈,采用最优网络混合策略设计了物理系统的控制器。这两个博弈过程相互融合,形成了一个基于博弈的安全控制研究方法。2.针对未知信息状态转移矩阵的CPS在DoS攻击下的安全问题,提出了基于博弈和基于强化学习的弹性控制算法。为了节约通信资源,在CPS中引入了事件触发机制。考虑到信息状态转移概率未知的情况,采用了基于在策略的SARSA方法来更新信息层的辅助矩阵。在上述条件下研究信息系统与物理系统间的关系,提出了一种融合博弈论、鲁棒控制理论、事件触发控制方法和SARSA学习方法的弹性控制算法,设计了弹性控制器和弹性控制策略来容忍一种具有概率特征的DoS攻击信号。3.针对一类线性时变CPS在混合攻击下的控制问题,提出了有限时域H_?控制器的设计算法。上述提到的混合攻击,包括对传感器和控制器之间通信通道发起的DoS攻击和对传感器和执行器发起的FDI攻击。考虑到混合攻击对CPS的物理性能可能造成更严重的危害,研究了攻击注入信号和被控输出的关系,来设计控制器增益以使闭环系统在有限时域内满足H_?性能条件。与此同时,为了减少最坏情况下的攻击输入信号对线性二次高斯性能的影响,通过求解一定条件下的两个耦合的倒向递推Riccati差分方程,建立了所设计控制器存在的充分条件。4.针对非线性CPS在混合攻击下的控制问题,提出了基于增益调度的安全控制器设计方法。这里的混合攻击,包括阻止测量数据或控制数据传输的DoS攻击和破坏传感器数据或执行器数据的FDI攻击。同时,考虑了物理过程中随机出现的扇形非线性,利用线性矩阵不等式工具设计了一种概率相关的增益调度控制器求解方法。由于出现非线性的实时概率未知的问题,将无限个线性矩阵不等式条件转化为有限个的形式,并引入松弛矩阵进行求解。最后对全文进行了总结,并提出了下一步的研究工作。更多还原

【Abstract】 Cyber physical system is a concept put forward in recent years.It refers to the system formed by the close integration of computing,communication and physical processes.Such a system can embed sensing,communication and computing technologies into physical devices,fully realizing distributed sensing,information transmission,intelligent information processing,real-time feedback and control.Hence,cyber physical system has been used in many fields,including chemical industry,grid power,water distribution,automobile and intelligent home.Information transmission is the most basic function in the cyber physical system,and the wired or wireless transmission equipment will be affected by all kinds of attacks from the cyber space,then causing the error or failure of information transmission,even causing the decline of control performance or the unstable operation of the system.Therefore,the security issueof cyber physical system has attracted more and more attention,and related research results have been extended from the initial communication field to the control field.The security issues of cyber physical system mainly fall into two categories:information security and security control theory.The former focuses on encryption and data processing;the latter studies how the network attacks affect the physical dynamics of control systems.The research of information security can’t prevent the attacker from affecting the control system,so the research of security control theory is necessary.The security control theory is based on the attack model to carry out the research,how to establish the appropriate attack model is very important.In previous studies,there are two most common types of attacks.The first type is denial of service attack,the other is spoofing attack,of which the false data injection attack is the typical representative.Recently,more and more attention has been paid to the security control problem of cyber physical system under these two types of network attacks,and some progress has been made,but there are still many shortages.For example,the existing researches of resilient control for the cyber physical system under DoS attack have not considering the case that the transition probabilities of cyber layer state are unknown,and the researches on the attack action are also relatively single,that is to say,most of which only consider the control problems for CPS under only one type of attack,furthermore,they are also lack of considering the influence of introducing event trigger and nonlinear sector,etc.Aimed at these shortages,this paper is to research the control system from the aspects of the stable operation,performance optimization and practical application,the concrete contents of which are as follows:1.A control strategy algorithm based on game is proposed for the security problem of CPS under DoS attack.On the one hand,considering the zero sum game of control performance,the hybrid strategy of cyber layer is designed by using physical performance index.On the other hand,considering the zero-sum stochastic game of attack-defense strategy,the controller of physical system is designed by using the optimal cyber hybrid strategy.These two game processes merge with each other to form a research method of security control based on game.2.A resilient control algorithm based on game and reinforcement learning is proposed for the security problem of CPS with unknown cyber state transition matrix under DoS attack.In order to save communication resources,event triggering mechanism is introduced in CPS.Considering the cyber state transition probability is unknown,an on-policy based SARSA method is adopted to update the auxiliary matrix of the cyber layer.Under the above conditions,the relationship between cyber system and physical system is studied,a resilient control algorithm combining game theory,robust control theory,event triggered control method and SARSA learning method is proposed,and a resilient controller and a resilient control strategy are designed to tolerate a series of DoS attack signals with probability characteristics.3.A design algorithm of finite horizon H_?controller is proposed for a class of linear time-varying CPS under hybrid attacks.The above-mentioned hybrid attacks,include the DoS attack on the communication channel between the sensor and controller,and the FDI attack on the sensors and actuators.Considering the hybrid attacks on the physical performance of the CPS may cause more serious harm,the relationship between the injected attack signals and the controlled output is studied,so that the H_?controller gain of the closed-loop system is designed to satisfy the performance condition in a finite horizon.At the same time,in order to reduce the impact of the injected attack signals in the worst case on the linear quadratic Gaussian performance,the sufficient conditions for the existence of the designed controllers are built up by solving two coupled backward recursive Riccati difference equations under certain conditions.4.A security controller design method based on gain scheduling is proposed for the control problem of nonlinear CPS under hybrid attacks.The hybrid attacks here include DoS attacks that block transmission of measurement data or control data,and FDI attacks that destroy sensor data or actuator data.At the same time,a probability-dependent gain scheduling controller is designed by taking into account the randomly occurring nonlinearities in the physical process.Because of the problem that the real-time probability of nonlinear is unknown,the infinite linear matrix inequation conditions are needed to be transformed into the finite linear matrix inequation conditions,and the slack matrix is introduced to solve the problem.Finally,the conclusion of the whole paper is given,and the next research work is put forward.更多还原