面向传感器攻击的概率时间窗感知融合算法研究

陈彦峰; 邓庆绪; 张天宇; 孙磊

doi:10.11897/SP.J.1016.2023.01227

TP393; 信息物理系统需要部署传感器对真实物理状态进行测量,通过网络传输到控制器实现各类功能.传感器攻击使得控制器接收的物理状态测量值存在较大误差,不能准确反映真实物理状态,从而促使控制器生成错误的决策和控制指令.除传感器攻击外,物理状态估测的准确性还受到两个因素影响:传感器测量过程中存在随机噪声,测量值在网络传输过程中存在随机延迟.针对以上因素,本文设计了能够抵抗传感器攻击的概率时间窗感知融合算法,对随机延迟进行补偿并计算在时间窗内不同传感器的正确概率累积值,进而作为权重调整卡尔曼滤波更新值,减小受攻击的传感器测量值带来的负面影响.自动驾驶车队仿真结果表明,在噪声和延迟的概率信息准确的条件下,与传统卡尔曼滤波、欧拉卡尔曼滤波、间隔融合等算法相比,本文所提出的延迟补偿-概率时间窗卡尔曼滤波算法具有最小的测量误差,物理状态估测累积误差较传统卡尔曼滤波降低67％;即使攻击者完全掌握本文融合算法对传感器进行自适应攻击,本文算法能够将融合值误差控制在可接受范围,系统仍能正常工作. Cyber-physical systems deploy sensors to measure the real physical state, and transmit their measurements to the controller through the network to achieve multiple functions. The sensor attack causes a large error in the physical state measurement value received by the controller, which cannot accurately reflect the real physical state, thus prompting the controller to make wrong decisions and control instructions. This directly affects the normal work of the system and threatens the related industrial production, personal and property safety. The security of cyber-physical systems is facing the increasingly serious problem caused by sensor attacks, and the research on related anti-attack algorithms is of great significance. This paper mainly focuses on sensor attacks, and studies how to improve the accuracy of physical state estimation through sensor fusion algorithm. Besides sensor attacks, there are two factors that affect the accuracy of physical state estimation: random noise in the sensing, and random delays in the network transmission. In view of the above factors, this paper designs a probability time window sensor fusion algorithm which can resist sensor attacks. Firstly, on the basis of using Kalman filter to deal with random noise, the corresponding state estimation value and probability are calculated by compensating the random communication delay. Then, the probability cumulative value of each sensor in the time window is used to evaluate the confidence of the normal operation of the sensor. Finally, the normal working confidence of each sensor is taken as the weight, and the physical state estimation value is updated by Kalman filter to reduce the negative impact brought by the measured value of the attacked sensor. The autonomous vehicle platoon MATLAB simulation results show that under the condition that the probability information of noise and delay is accurate, the algorithm proposed in this paper has the lowest sensing error, compared with ordinary Kalman filter, Euler-Distance Kalman filter and Interval Fusion. And the cumulative error of physical state estimation is 67% lower than that of ordinary Kalman filter. In addition, we evaluate the performance of the proposed algorithm under inaccurate probability information of noise and delay with different degrees of error, including the range and the parameters of distribution functions. The results show that all relative algorithms need relatively accurate information to generate acceptable estimation of the physical state to make the autonomous vehicle platoon work normally. The proposed algorithm can tolerate the maximum error of the probability information among the above methods. When the relative error between the communication delay time range obtained by the controller and the real range is +60%, the proposed algorithm can still achieve higher fusion accuracy than the traditional Kalman filter algorithm. If the attacker obtains the full-knowledge of the proposed method and launches adaptive sensor attacks, it will pre-check whether the designed attack vector can be detected by the proposed algorithm. To keep stealthy, the fused errors are limited to acceptable range under which the attacked sensor will still be regarded as normal sensor. In this condition, the maximum negative affect the attacker can produce is the fused value skipping between extreme points of the acceptable range, under which the system can still work normally.

面向传感器攻击的概率时间窗感知融合算法研究

View Online

Abstract

Details

Metrics