Proposing an Islanding Detection Index for Distribution Networks with Conventional and Inverter-Based Distributed Generators

This paper proposes an index based on zero-sequence components that serves as a single indicator for islanding detection. The proposed index demonstrates satisfactory performance, achieving an accuracy of 97.84% using a simple threshold adjustment model. Additionally, this index can be integrated into artificial intelligence-based methods to enhance accuracy further. The results indicate that utilizing the proposed index as a single indicator within a one-dimensional convolutional neural network (1D-CNN) model yields a competitive accuracy of 99.78%. This outcome is noteworthy when compared to more advanced artificial intelligence methods, such as long short-term memory (LSTM) neural networks, which rely on a larger set of features. The data collection for testing encompasses various islanding and non-islanding conditions, including islanding under different loading conditions and power factors, as well as varying quality factor values. Non-islanding scenarios include the activation and deactivation of large loads and capacitor banks, along with the application of various short-circuit faults at different locations and resistances. The results from all tests demonstrate the superiority of the proposed method and index.