CAREER: PARP: Mislead Physical-Disruption Attacks by Preemptive Anti-Reconnaissance for Power Grids Cyber-Physical Infrastructures

与一般计算环境中的攻击相比，针对电网等工业控制系统的网络攻击具有独特的特征。由于物理过程在复杂的物理模型下运行，因此攻击者在渗透内部控制网络后，会对网络和物理基础设施进行深入侦察。全面的侦察使它们能够造成快速和可能不可逆转的破坏，如停电，经济损失，甚至人员伤亡。这个CAREER项目旨在设计和量化先发制人的反侦察技术，这些技术将误导对手关于电网的网络物理基础设施。为了达到防止物质损害的目标，该项目将带来明显的好处。首先，它会在恶意活动发起之前误导攻击，提前消除潜在威胁，从而防止损害。其次，防止对一组关键物理数据进行侦察可以防止各种攻击，包括未知攻击。为了实现这一目标，该项目将推进现有的反侦察方法与两个关键的研究推力。第一个研究重点旨在通过原始控制功能虚拟化来误导对手对网络基础设施的看法。利用软件定义网络所支持的网络可编程性，该技术虚拟化了在电网中提供控制功能的网络通信，而无需为每个物理设备构建虚拟机。在保持控制功能的性能的同时，虚拟化中和来自各种物理设备的通信模式，并基于中和的模式欺骗节点。因此，这从通信信道中删除了特定于设备的功能，并通过欺骗节点提供误导性信息，防止对手将易受攻击的设备定位为目标。第二个研究重点是通过创建一个电子模型引导的生成对抗网络来误导对手的物理基础设施。具体来说，这种方法通过将电气模型集成到其内部结构中并制作生成真实的和诱饵数据的电网来推进当前生成对抗网络的设计。真实的和诱饵数据的组合将符合电气模型，但呈现出与实际电网不同的画面。因此，诱饵误导对手设计无效的攻击策略，以避免损害实际的网格，而真实的数据继续服务于合法的控制应用程序。研究推力将在一个硬件在环电力系统测试平台上进行评估，该项目由安全可信网络空间（SaTC）计划和激励竞争研究（EPSCoR）计划共同资助，该奖项反映了NSF的法定使命，并通过使用基金会的知识价值和更广泛的影响审查标准进行评估，被认为值得支持。

Cyberattacks targeting industrial control systems like power grids present unique features compared to attacks in general computing environments. Because physical processes operate under complicated physical models, adversaries perform in-depth reconnaissance on both cyber and physical infrastructures after penetrating internal control networks. The comprehensive reconnaissance enables them to cause rapid and possibly irreversible damage like power outages, economic losses, and even human casualties. This CAREER project aims to design and quantify preemptive anti-reconnaissance techniques that will mislead adversaries about power grids’ cyber-physical infrastructures. Toward the objective of preventing physical damage, this project will introduce clear benefits. First, it will mislead attacks before malicious activities are launched, removing potential threats in advance and thus preventing damage. Second, preventing reconnaissance on a critical set of physical data can protect against a wide spectrum of attacks, including unknown ones. To achieve this objective, this project will advance existing anti-reconnaissance approaches with two critical research thrusts. The first research thrust aims to mislead adversaries about cyber infrastructures by an original control function virtualization. Leveraging network programmability enabled by software-defined networking, this technique virtualizes network communications that deliver control functions in power grids without building virtual machines for each physical device. While preserving the performance of control functions, virtualization neutralizes communication patterns from various physical devices and spoof nodes based on the neutralized patterns. This consequently removes device-specific features from communication channels and delivers misleading information by the spoofed nodes, preventing adversaries from pinpointing vulnerable devices to target. The second research thrust seeks to mislead adversaries about physical infrastructure by creating an electrical-model-guided generative adversarial network. Specifically, this method advances the current design of generative adversarial networks by integrating electrical models into their internal structures and crafting power grids that generate real and decoy data. The combination of real and decoy data will conform to electrical models but present a different picture from the actual power grid. Consequently, decoys mislead adversaries into designing ineffective attack strategies that avert damage to the actual grid, while real data continues to serve legitimate control applications. The research thrusts will be evaluated in a hardware-in-the-loop power system testbed, which can integrate real-time operations of physical devices with high-fidelity simulators to quantify power system runtime reactions to external events.This project is jointly funded by the Secure and Trustworthy Cyberspace (SaTC) program and the Established Program to Stimulate Competitive Research (EPSCoR).This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.