Abstract
Unmanned ground vehicles (UGVs) are vulnerable to stealthy false data injection (SFDI) attacks, which compromise system integrity by altering control signals and injecting fraudulent sensor data. Most existing methods for detecting SFDI attacks rely on computationally intensive approaches and overlook the unified design of attack detection and state estimation frameworks. To address these limitations, this study proposes a lightweight protection framework that integrates detection and state estimation for SFDI attacks. The proposed framework encodes the system's internal state, eliminating the need to transmit encoded information seeds and thus reducing the risk of compromise. Reinforcement Q-learning is employed to enhance detection performance against micro-amplitude SFDI attacks. Additionally, an attack-resilient state estimator with plug-and-play capabilities is developed using Kalman filtering, achieving an integrated framework for attack detection and state estimation. The effectiveness of the proposed method is validated experimentally using a UGV as the test platform and compared with related methods. Experimental results demonstrate that the proposed method exhibits strong generalizability and low computational complexity. It achieves a 100% detection rate for various types of SFDI attacks, particularly those with small magnitudes, and maintains an average state estimation accuracy of 95%. Moreover, the framework proves resilient to computational delays of up to 200 ms and 30 ms for detection and estimation, respectively, showcasing its practical applicability and robustness in real-world scenarios.
Funding Information
  • National Natural Science Foundation of China (52332011)
  • National Natural Science Foundation of China (U2241214)

This publication has 37 references indexed in Scilit: