A Machine Learning Approach toward Cyber-Physical Security of Battery Systems

Abstract

Modern battery systems exhibit a cyber-physical nature due to the extensive use of communication technologies in battery management. This makes modern cyber-physical battery systems vulnerable to cyber threats where an adversary can manipulate sensing and actuation signals to satisfy certain malicious objectives. In this work, we present a machine learning-based approach to enable resilience to adversarial attacks by detecting and estimating the attack and subsequently taking corrective action to mitigate the attack. In particular, we focus on false data injection type attacks on battery systems. The overall diagnostic algorithm consists of an adaptive boosting-based attack detector, a long short-term memory (LSTM) neural network-based attack estimator, and a corrective controller. The proposed algorithm is trained and tested by utilizing data from a complex battery electrochemical battery simulator. Simulation results are presented to verify the effectiveness of the approach.