Investigation of Cell Voltage Buffer Manipulation Attack in a Battery Management System using Unsupervised Learning Technique

Allowable voltage difference (buffer) between cells in a battery module is defined in the balancing algorithm of the battery management system (BMS) by the manufacturer and can be updated to reduce or increase the balancing time. When this buffer is slowly increased, the battery module faces degradation which is undetectable unless it exceeds the safe operation threshold, at which point it is detected either by the capacitor bank on the power conversion unit (PCS) side or by the system controller. This parameter could thus be exposed to an attack, which if prolonged for a longer duration, can ultimately reduce the overall lifetime of the battery. In this paper, a BMS algorithm is simulated under normal and manipulated voltage buffer values. To effectively detect the voltage and state of charge (SOC) convergence (balancing) periods, a principal component analysis (PCA) based unsupervised k-means approach is applied on the voltage and SOC datasets and compared with a standalone k-means approach. The results suggest that the PCA based k-means approach outperforms the standalone approach.