Islanding Detection Using Transformer Neural Networks

Unintentional islanding is becoming a significant concern for distribution utilities due to the growing use of solar photovoltaic systems. These unintentional islands pose a threat to the safety of workers and sensitive equipment with unplanned energization and poor power quality. Traditional detection methods have been shown to have non-detection zones. Emerging machine learning-based methods for detecting unintentional islands are complex to model, as they require manual feature extraction from the signals. We propose an end-to-end solution by using Time Series Transformers, which simplifies the modeling process by automating the feature extraction stage. Our results show that the Time Series Transformer for islanding detection outperforms other machine learning methods, offering a simple but reliable detection model.