An open-source tool for simulating cyberattacks on smart grid datasets

Data-driven approaches based on machine learning and deep learning are effective countermeasures for identifying attacks in cyber–physical system infrastructures. However, performance evaluation requires the presence of attacks in datasets. This paper presents two open-source Python tools for simulating cyber attacks. The first, called the Unpoisoned Attack Scenario, simulates attacks in benign training and testing sets to support supervised training and evaluation of supervised and unsupervised models. The second, called the Projected Gradient Descent (PGD)-Based Adversarial Sample Generation Scenario, generates adversarial samples in labeled and unlabeled training sets for adversarial training. Although developed for smart grid datasets, the tools can be applied to other numerical time-series datasets.