Detecting False Data Injection in Smart Grid Overcurrent Relay Settings Using an AE–GAN–SVM Framework

ABSTRACT

Cybersecurity incidents in power grids have surged by over 60% in the past decade, revealing the vulnerability of protection layers to false data injection (FDI) attacks. Such attacks compromise phasor measurement units (PMUs) and overcurrent relays by altering measurement streams or relay configuration settings. This paper proposes a hybrid Autoencoder–Generative Adversarial Network–Support Vector Machine (AE–GAN–SVM) framework for online detection of coordinated FDI attacks in smart grid environments. The GAN generates realistic synthetic data under multiple fault and attack scenarios to mitigate dataset scarcity, while the autoencoder learns latent nonlinear representations from the augmented space. The resulting features are classified by an SVM to discriminate authentic from falsified samples. Unlike conventional techniques that isolate measurement‐ or setting‐level anomalies, the proposed method jointly monitors both domains, enabling robust cyber‐physical security coverage. Simulation results on the IEEE 14‐Bus test system demonstrate an average detection accuracy of 99.17%, outperforming existing approaches across various fault conditions—single‐phase‐to‐ground, double‐phase‐to‐ground, three‐phase‐to‐ground, double‐phase and three‐phase faults. The results confirm the model's capability to safeguard relay coordination integrity and enhance the cyber‐resilience of smart grid infrastructures.