Generative Adversarial Networks-Based False Data Injection: A Concern for Data Integrity of Power Networks

The digitalisation of power grids has increased data integrity and security risks, enabling subtle manipulations to cause power theft, false tripping, and compromised controls. Effective false data attacks must exhibit complex nonlinear behaviour, preserve network dependencies, and account for real-time system conditions to evade advanced false data detection (FDD) methods. Constructing such attacks is highly challenging due to the restricted access to critical system information. In this paper, three deep learning-based False Data Injection (FDI) attacks for power networks are proposed, highlighting the vulnerabilities of existing false data detection defences. A holistic comparative analysis of attack constructed by three variants of Generative Adversarial Networks (GANs), including GAN, Wasserstein GAN (WGAN) and Conditional GAN (CGAN), is presented. The proposed false data attacks are assessed against five diverse FDD defences to assess all factors of possible FDI attack failure. IEEE Case-5, Case-14, Case-30 and Case-118 bus systems are simulated as target networks with realistic demand modelling. The simulation results display a progressive improvement of FDI success from GAN to CGAN.