Flexible DC transmission lines protection method based on StarNet

To address the challenges of threshold selection in existing fault diagnosis methods for flexible DC transmission lines, this study proposes a fault identification approach integrating gramian angular field (GAF) and StarNet-based analysis. The methodology initiates by transforming sequential voltage data under diverse fault conditions into two-dimensional voltage GAF images through GAF-based encoding. These generated voltage GAF images are subsequently partitioned into training and testing datasets to facilitate the training and validation processes of the StarNet. During model development, the network architecture is optimized through partial convolutional layers, while convergence efficiency is enhanced via a cosine annealing algorithm. Comprehensive simulations evaluating transition resistance, noise interference, sampling frequency, and lightning strike disturbances demonstrate the proposed method’s adaptability and superior diagnostic performance across multiple operational scenarios.