Online Dictionary Learning Based Fault and Cyber Attack Detection for Power Systems
This addresses cybersecurity risks in power grid monitoring systems, though it appears incremental as it builds on existing dictionary learning and online classification techniques.
The paper tackles the problem of distinguishing cyber-attacks from normal disturbances in power systems by using an online dictionary learning approach with a Hoeffding adaptive tree classifier, achieving superior performance over state-of-the-art methods in simulations on a modified IEEE 9-bus system.
The emerging wide area monitoring systems (WAMS) have brought significant improvements in electric grids' situational awareness. However, the newly introduced system can potentially increase the risk of cyber-attacks, which may be disguised as normal physical disturbances. This paper deals with the event and intrusion detection problem by leveraging a stream data mining classifier (Hoeffding adaptive tree) with semi-supervised learning techniques to distinguish cyber-attacks from regular system perturbations accurately. First, our proposed approach builds a dictionary by learning higher-level features from unlabeled data. Then, the labeled data are represented as sparse linear combinations of learned dictionary atoms. We capitalize on those sparse codes to train the online classifier along with efficient change detectors. We conduct numerical experiments with industrial control systems cyber-attack datasets. We consider five different scenarios: short-circuit faults, line maintenance, remote tripping command injection, relay setting change, as well as false data injection. The data are generated based on a modified IEEE 9-bus system. Simulation results show that our proposed approach outperforms the state-of-the-art method.