Active Learning-Based XGBoost for Cyber Physical System Against Generic AC False Data Injection Attacks
Wenli Xue, Ting Wu
Abstract
With the exponential growth of information and communication technology, the traditional power system is gradually evolving into a cyber physical energy system (CPES) with frequently interactions between physical and cyber components. CPES indeed revolutionize the power grid efficiency and operational performance, but it also gives rise to new security challenges, causing catastrophic consequences in power system. Therefore, this article proposes a new false data injection attack (FDIA) detection mechanism to automatically recognize the intrusions and thereby improve the cybersecurity of CPES. In this mechanism, a two-stage FDIA model is developed to generate training dataset for the machine learning-based detector, and an extreme gradient boosting (XGBoost) classifier is meticulously designed by combining with active learning and Bayesian optimization techniques to improve training efficiency and model performance respectively. The proposed FDIA defense strategy can properly extract the nonstationary and nonlinear attacking features and thus locate the biased states with a high accuracy. The obtained numerical results on the standard IEEE 14-, 57- and 118-bus systems highlight the destructive effects of the developed two-stage FDIA and comprehensively demonstrate the effectiveness of the proposed FDIA detection mechanism.