Lee Chanhwa
School of Intelligent Mechatronics Engineering, Sejong University, Seoul 05006, Korea.
Sensors (Basel). 2022 Sep 13;22(18):6909. doi: 10.3390/s22186909.
This paper considers a discrete-time linear time invariant system in the presence of Gaussian disturbances/noises and sparse sensor attacks. First, we propose an optimal decentralized multi-sensor information fusion Kalman filter based on the observability decomposition when there is no sensor attack. The proposed decentralized Kalman filter deploys a bank of local observers who utilize their own single sensor information and generate the state estimate for the observable subspace. In the absence of an attack, the state estimate achieves the minimum variance, and the computational process does not suffer from the divergent error covariance matrix. Second, the decentralized Kalman filter method is applied in the presence of sparse sensor attacks as well as Gaussian disturbances/noises. Based on the redundant observability, an attack detection scheme by the χ2 test and a resilient state estimation algorithm by the maximum likelihood decision rule among multiple hypotheses, are presented. The secure state estimation algorithm finally produces a state estimate that is most likely to have minimum variance with an unbiased mean. Simulation results on a motor controlled multiple torsion system are provided to validate the effectiveness of the proposed algorithm.
本文考虑了存在高斯干扰/噪声和稀疏传感器攻击情况下的离散时间线性时不变系统。首先，我们提出了一种基于可观测性分解的最优分散多传感器信息融合卡尔曼滤波器，用于无传感器攻击的情况。所提出的分散卡尔曼滤波器部署了一组本地观测器，它们利用自身的单个传感器信息并为可观测子空间生成状态估计。在无攻击情况下，状态估计实现最小方差，并且计算过程不会受到发散误差协方差矩阵的影响。其次，将分散卡尔曼滤波器方法应用于存在稀疏传感器攻击以及高斯干扰/噪声的情况。基于冗余可观测性，提出了一种通过χ2检验的攻击检测方案以及一种在多个假设之间基于最大似然决策规则的弹性状态估计算法。安全状态估计算法最终产生一个最有可能具有最小方差且均值无偏的状态估计。给出了在电机控制多扭转系统上的仿真结果，以验证所提算法的有效性。
