An Effective Hybrid Approach for Detection of False Data Injection Attacks in Connected Battery Systems with Noisy Measurements

In this paper, an effective method based on adaptive extended Kalman filter (AEKF) is proposed for detection of random FDIs against battery state-of-charge algorithms on cloud battery management platforms. First, the battery model is established and used with the AEKF to predict the battery response. Second, a residual signal (RS) is defined as the difference between the AEKF-based estimated battery voltage and the received voltage measurement. The FDIs are then detected based on a hybrid detection criterion mixing the Chi-squared test and Euclidean detector. The proposed mixed strategy improves the detection accuracy in terms of false negatives and false positives caused by noises and changes in battery operation. Regarding the latter point, the AEKF is equipped with a dedicated recursive least squares filter to accommodate real-time model changes. The proposed algorithm is developed and verified based on actual battery data related to high-capacity lithium-ion cells. The method is exposed to different case studies considering normal and attack conditions and a remarkable detection accuracy of about 98% is attained with no false positive in the presence of current and voltage noises up to ±10 mA and ± 3 mV.