信息物理系统的分布式融合攻击辨识理论与方法研究

With more concern on the security problem of cyber-physical systems, the attack identification problem has aroused the researchers' attention rapidly. The existing attack identification methods are mostly based on local measurement information driven centralized strategies, as a result of severe lack of measurement redundancies, significant deficiencies can be found in identification precision and resilience. The cyber-physical systems are generally large-scale systems, which have multiple sensors and huge measurement redundancies. Therefore, this project aims to develop the distributed fusion attack identification methods for cases with combinatorial attacks, high-intensity attack and stealthy attack. Specifically, the proposed methods will improve the identification precision by optimizing the local attack estimations via distributed fusion mechanism, meanwhile, for cases with high-intensity attack and stealthy attack, the resilience of fusion attack identification mechanism will also be guaranteed by data scheduling strategies, then the proposed methods will be applied to the experiment platform of localization and tracking system of mobile robot. Finally, all the aforementioned designs make up an effective systematic distributed multi-sensor attack identification methodology. The project will improve the reliability of the existing attack identification methods, and also boost the development of the defense theory of cyber-physical systems.

随着信息物理系统（cyber-physical systems, CPS）安全问题日趋突出，攻击辨识问题迅速引起研究者关注。现有攻击辨识方法均是基于单一局部测量信息的集中式攻击辨识策略，由于测量信息冗余度极为有限，在辨识精度和辨识机制弹性两方面均存在明显不足。CPS通常是包含多节点多子系统的大规模系统，具有大量传感器冗余测量信息。为此，本项目结合多传感器信息融合技术，分别研究组合攻击、强攻击、以及隐蔽攻击情况下的分布式融合攻击辨识问题，通过对各局部攻击估计值进行进一步融合以提高攻击辨识精度，同时通过数据调度策略确保融合攻击辨识机制在强攻击或隐蔽攻击下的弹性，并将所提算法应用于移动机器人目标定位与跟踪系统实验平台，最终形成一套有效的分布式融合攻击辨识理论与方法。本项目将有效提高现有攻击辨识技术的可靠性，并促进CPS安全防御理论的发展与完善。

信息物理系统安全问题日趋突出，然而，现有攻击辨识方法在辨识精度和辨识机制弹性两方面均存在明显不足。本项目首先针对传感器攻击下的CPS，通过设计卡尔曼融合状态估计器提高了系统状态辨识精度，进而设计了基于蛮力搜索的多传感器在线切换机制确保了状态观测器的弹性，解决了切换攻击下的状态估计问题；其次，针对传感器攻击下的多智能体系统，提出了基于通信拓扑搜素的弹性切换状态观测器设计；同时，研究了一类连续时间多智能体系统的分布式攻击辨识问题，考虑在攻击频率突变情况下，提出了带有一维增益动态调节机制的中间观测器；在此基础上，针对离散时间CPS提出了带有二维动态增益调节能力的中间观测器，进一步提升了中间观测器的攻击辨识弹性和准确性。最后，针对绝大多数观测器在辨识高频攻击信号时过渡性能差、精度不足等问题，我们提出了基于迭代学习的智能化中间观测器设计，彻底去除了中间观测器对于线性矩阵不等式方法的依赖，实现低时延、低误差攻击辨识，有效提升了基于中间观测器的攻击辨识方法在伺服电机、移动机器人、工业机械臂等实际系统中的应用价值。