CAREER: A dynamic game theoretic approach to cyber-security of controlled systems

随着基础设施和工业控制系统越来越多地连接到互联网和其他可广泛访问的网络，它们对网络攻击的脆弱性正迅速成为一个日益严重的问题和国家优先事项。成功抵御这些攻击需要严格的安全措施，反过来，对控制算法提出了独特的要求，这些算法必须能够在存在恶意和战略入侵的情况下确保一定程度的闭环稳定性和性能，并补充传统网络安全技术提供的第一道防线。为了在今天能够实施？sSCADA系统，这些算法必须（1）能够面对各种各样的攻击，（2）易于部署，需要相对较低的计算资源，（3）至少对某些定义明确的攻击类别具有弹性的理论保证。这个CAREER项目提出使用博弈论工具，更具体地说是动态零和博弈，停止游戏和团队理论这种建模范式所提供的灵活性使得通过攻击者来捕获攻击的各种特征成为可能。s和控制器？的信息和动作集，以及通过游戏的内核。除了易于实施的控制策略，这种博弈论的方法也提供了基本的见解的影响，攻击？的切入点和其结果的隐蔽性，并指向具体的弹性架构的网络控制系统。由此产生的算法将在三个实验测试平台上实现和验证，这些测试平台的复杂性和真实性不断增加。其中一个测试平台是位于瑞典斯德哥尔摩的一个实际的小规模SCADA电网，PI将通过与皇家理工学院（KTH）的持续合作获得访问权限。与该研究计划紧密结合的是教育和推广活动，这些活动将部分利用测试平台的实施，旨在扩大博弈论工具在网络安全社区和工程专业学生中的使用，并分别为公众揭开网络攻击的神秘面纱。所提出的设计可证明安全的控制系统的方法的新奇和针对性在于它关注控制算法本身的弹性，并结合动态博弈论和鲁棒控制技术的互补优势来实现这一目标。这与传统的网络安全方法形成鲜明对比，传统的网络安全方法通常关注减少软件和硬件漏洞，并在入侵点阻止攻击。此外，虽然博弈论方法已被证明在计算机和通信网络环境中的安全问题的制定和解决方面是成功的，但使用这些工具来研究和设计网络攻击下的安全网络控制系统是相对较新的和未探索的。我们希望从这个项目产生的算法和方法是广泛适用的，并提供额外的选择？纵深防御？网络控制系统的策略。为了将这些结果传播给广大的学生，研究人员和从业人员，每年的暑期学校？决策与安全的博弈论将作为一个独立的事件，或作为UIUC的一部分组织？信息信任研究所？s（ITI）提供专门的夏季研讨会。本科生将通过ITI招聘？的暑期实习计划，并将参与两个内部测试平台的实施和表征。这些试验台还将用于对当地和全球公众的外联工作。最后，该项目将有助于加强UIUC和KTH之间的研究和教育联系，包括学生交流。

As infrastructure and industrial control systems are increasingly being connected to the Internet and other widely accessible networks, their vulnerability to cyber attacks is fast becoming a growing concern and a national priority. Successfully weathering these attacks calls for stringent security measures and, in turn, imposes unique requirements on control algorithms, which must be able to ensure some level of closed-loop stability and performance in the presence of malicious and strategic intrusions, and complement the first line of defense provided by more traditional cyber-security techniques. In order to be implementable on today?sSCADA systems, these algorithms must (1) be able to face a wide variety of attacks, (2) be easy to deploy and require relatively low computational resources and, (3) come with theoretical guarantees of resilience for at least some well-defined classes of attacks.This CAREER project proposes to explore the design of such secure control algorithms using the tools of Game Theory, more specifically dynamic zero-sum games, stopping games, and team theory. The flexibility afforded by this modeling paradigm makes it possible to capture various characteristics of attacks through the attacker?s and controller?s information and action sets, as well as through the kernel of the game. In addition to readily implementable control strategies, this game theoretic approach also provides fundamental insights into the influence of an attack?s point of entry and stealthiness on its outcomes, and points towards specific resilient architectures for networked control systems. The resulting algorithms will be implemented and validated on three experimental testbeds of increasing complexity and degree of realism. One of these testbed is an actual small scale SCADA-enabled power grid located in Stockholm, Sweden, which the PI will be granted access to via ongoing collaboration with the Royal Institute of Technology (KTH). Tightly integrated with this research plan are education and outreach activities, which will partly leverage the testbed implementations, and are aimed at extending the use of game theoretic tools in the cyber-security community and among engineering students, and demystifying cyber-attacks for the general public, respectively.Intellectual Merit: The novelty and pertinence of the proposed approach for designing provably secure control systems lies in its focus on the resilience of the control algorithm itself, and its combination of the complementary strengths of dynamic game theory and robust control techniques to achieve this goal. This is in contrast with traditional cyber-security approaches, which are typically concerned with decreasing software and hardware vulnerabilities, and thwarting attacks at their point of entry. In addition, while game theoretic methods have proved successful in the formulation and resolution of security problems in the context of computer and communication networks, the use of these tools for studying and designing secure networked control systems under cyber-attacks is relatively new and unexplored.Broader Impact: We expect the algorithms and approach resulting from this project to be widely applicable, and to provide additional options for the design of ?defense in depth? strategies for networked control systems. In order to disseminate these results to a broad community of students, researchers, and practitioners, a yearly summer school on ?Game Theory for Decision-Making and Security? will be organized either as a stand-alone event, or as part of UIUC?s Information Trust Institute?s (ITI) offering of dedicated summer workshops. Undergraduates will be recruited through ITI?s Summer Internship Program, and will be involved in the implementation and characterization of the two in-house testbeds. These testbeds will also be used in outreach efforts to the local and global public. Finally, this project will help strengthen the research and educational ties between UIUC and KTH, including student exchange.