Leveraging High-Fidelity Datasets for Machine Learning-based Anomaly Detection in Smart Grids

Data-driven anomaly detection systems are increasingly becoming essential for protecting critical cyber-physical system (CPS) infrastructure, such as the power grid, against the growing number of sophisticated cyber-attacks. The development of such tools is reliant on the availability of high-fidelity cyber-physical datasets that cover a diverse variety of potential cyber events. In this work, a co-simulation smart grid platform is utilized to develop a realistic dataset, which is used to train and test a machine learning-based anomaly detection system (ADS). The evaluation of the developed ADS shows robust performance even when tested with statistically diverse test data not used in training. This work is a preliminary step towards the development of a cyber-resilient middleware framework, which will serve as a testbed for the development and evaluation of cybersecurity solutions and CPS datasets.