Comparative Analysis of Tabular Generative Adversarial Network (GAN) Models for Generation and Validation of Power Grid Synthetic Datasets

The demand for securing SCADA (Supervisory Control and Data Acquisition)-based power grid systems from cyber-attacks has been increasing significantly in the last few years. Current research trends widely adopt Machine Learning (ML) techniques to prevent attacks against such critical infrastructure. However, the efficiency of these techniques largely depends upon the availability of large datasets. Acquiring large data from such critical systems is not always feasible and this has inhibited the research progress in the development of advanced ML algorithms that can make a notable difference in the prediction of malicious events. Thus, there is a strong need for generating large synthetic yet realistic datasets from existing small datasets. This paper presents a comparative analysis of tabular Generative Adversarial Network (GAN) models for the generation and validation of synthetic datasets from existing datasets of power grids. Moreover, the synthetic datasets are validated using statistical analysis, and machine learning efficacy. These synthetic datasets open opportunities for the research community to explore advanced machine learning and deep learning methodologies for the protection of industrial systems.