False Data Injection Attack (FDIA) Detection

Overview

False Data Injection Attacks (FDIAs) in smart grid systems can evade traditional bad-data detection algorithms, potentially causing widespread power outages. This project designs a hybrid deep learning pipeline combining multiple architectures for robust FDIA detection.

Architecture

The detection pipeline combines:

• Autoencoder (AE): Unsupervised pre-screening of anomalous measurement patterns
• Graph Neural Network (GNN): Topological analysis of power grid connectivity
• Recurrent Neural Network (RNN): Temporal modeling of measurement sequences
• Deep Belief Network (DBN): Feature extraction from raw sensor readings
• GAN-based augmentation: Synthetic attack sample generation for training data enrichment

Results

Achieves 98.6% detection rate with false alarm rate below 1.2% on the IEEE 14-bus and 118-bus test cases.