Prediction-Based Data Augmentation for Smart Grid Line Outage Detection

Line outage detection is an essential component of situational awareness and security of the smart grid. Transmission lines are an attractive target for advanced persistent threats, and their outages can be caused by cyber or physical attacks. In this paper, we demonstrate the effectiveness of prediction-based data augmentation in enhancing the detection and location of line outages. Moreover, we show how prediction-based data augmentation can benefit various data-driven learning techniques, specifically support vector machines, naive Bayes, and convolutional neural networks. This work presents the results of six evaluation cases using six training compositions to show that, on average, our enhanced method of prediction-based data augmentation can increase the accuracy of line outage detection by 4.19% compared to using only historical data for training. Furthermore, we test the algorithm under ten measurement availability conditions. These tests demonstrate our algorithm's effectiveness, particularly under limited data availability.