A Transfer Learning Based Graph Convolutional Network for Intelligent Load Shedding in Modern Power Systems for Unknown Faults

Deep learning-based event load shedding (ELS) methods have successfully prevented transient instability. However, in power systems, faults may occur outside the training database, reducing the model's adequate performance. On the other hand, updating the model for new faults requires an extensive database. To solve the issue of unknown faults, this paper presents a method based on the Graph Convolutional Network (GCN) utilizing transfer learning, enabling the model to be updated with limited data. An ELS model is trained with an extensive database in the first step. Then, if a new fault occurs, the model is transferred to the latest fault and updated using transfer learning and a small database. Evaluation of the GCN-based model on the IEEE 39 bus system shows its effective performance in determining the optimal LS.