Machine Learning Based Intrusion Detection System for Real-Time Smart Grid Security

The main objective of this paper is to develop an efficient, scalable, and faster machine learning (ML) based tool for real-time smart grid (SG) security. With the integration of information and communication technologies (ICT), power grid operations have become vulnerable to false data injection attacks. This paper presents an ML-based intrusion detection system (IDS) for smart grid security by developing an intelligent module that uses Principal Component Analysis (PCA), Linear Discriminant Analysis (LDA), and Support Vector Machine (SVM) in combination. A two-stage methodology has been proposed for the detection of false data injection attacks in the Smart Grid system. The first stage is responsible for data dimensionality reduction using PCA or LDA, followed by data classification using SVM in the second stage. The proposed intelligent module uses the real-time measurement data retrieved from the phasor measurement units (PMUs) which are assumed to be placed optimally in the power network for grid observability. Upon receiving a fault signal, the protection system checks on the incoming data patterns and compares them with the behaviour of system dynamics. With machine learning algorithms, the incoming fault signal is classified as an actual (real) fault or a false (fake) fault with malicious intentions. The proposed intrusion detection systems have been validated on the three buses and the benchmark IEEE 14 and IEEE 30 bus system.