Improved Wasserstein Generative Adversarial Networks Defense Method Against Data Integrity Attack on Smart Grid

Background: As the integration of communication networks with power systems is getting closer, the number of malicious attacks against the cyber-physical power system is increasing substantially. The data integrity attack can tamper with the measurement information collected by Supervisory Control and Data Acquisition (SCADA), which can affect important decisions of the power grid and pose a significant threat to normal operation. Objective: In order to defend against the data integrity attack, and safeguard the normal operation of the power system, we propose a defense method based on improved WGAN (Wasserstein Generative Adversarial Networks). Methods: Firstly, through the training of the discriminator and generator, the discriminator identifies and eliminates the measurements damaged by the data integrity attack, and the generator uses the temporal and spatial correlation of the power grid measurements to generate complementary measurements which are infinitely close to the original normal data. Meanwhile, using the improved WGAN, the training stability and the convergence speed are significantly improved, and the quality of complementary data is much higher. Results: Extensive simulation experiments were carried out in the IEEE-14 and IEEE-118 standard bus systems. By comparing the deviations between the complementary data and normal data under different iteration times and methods, the mean squared error of the deviations with the proposed method is greatly reduced than traditional methods. The excellent recovery performance of the defense method based on improved WGAN is verified. Conclusion: A smart grid data integrity attack defense method based on improved WGAN is proposed to detect data integrity attacks, and the experimental results demonstrate the recovery effectiveness of the proposed method.