A Sequence Component-Based Feature for Passive and Artificial-Intelligence-Based Islanding Detection

Feature selection is a critical aspect of designing islanding detection schemes, as it affects the method’s accuracy, speed, complexity, and overall performance. Various features are suggested in the literature and integrated into different (AI)-based methods to achieve high performance in islanding detection. In this paper, a feature based on electrical zero sequence components is proposed, which can be used as the sole feature for islanding detection. The proposed feature demonstrates favorable results of 97.84% when using a simple threshold model. Additionally, the proposed feature can be employed in AI-based methods to achieve high balanced accuracy.Based on the obtained results, using the proposed feature as the only feature in a 1D CNN model yields a competitive result of (99.78 ± 0.012). This feature is a key aspect when comparing the results of AI-based methods, such as sophisticated Long Short-Term Memory (LSTM) neural networks, that utilize a higher number of different features.