Boundary Generative Adversarial Network-Based Anomalous Traffic Detection for the Smart Grid Internet of Things

With the rapid growth of the Grid Internet of Things (GIT), new power business lacks device security in the initial stage. At present, IoT has relatively little attack traffic, while normal traffic dominates. Therefore, researchers tend to adopt anomaly detection methods to address the scenario. However, current anomaly detection methods are still incompetent in dealing with anomalous features in high-dimensional network data. Meanwhile, the anomaly score thresholds of the samples usually need to be set manually by subjectivity, which limits their effectiveness in practical applications. Therefore, we propose a boundary generative adversarial network-based for the Grid Internet of Things. By applying an adversarial learning method to generative adversarial networks, we aim to improve the model’s ability to learn features and thus train a discriminator that can accurately detect attack behaviors. This innovative approach dramatically improves the accuracy of attack behavior detection and also solves the challenges of supervised deep learning and unsupervised anomaly detection.