A Transfer Learning Method for Intelligent Load Shedding using Graph Convolutional Network considering Unknown Faults

Event-based load shedding (ELS) is a vital emergency countermeasure against transient voltage instability in power systems. Deep learning(DL)--based ELS has recently achieved promising results. However, in power systems, faults may occur that are not in the training database, reducing the model's effective performance. In this situation, it is necessary to update the model. On the other hand, updating the model for new faults requires a large database. To address the problem of unknown faults, this paper proposes a transfer learning-based graph convolutional network (GCN) model that allows updating the model with a small database. In the first step, an ELS model is trained with a large database. Then, if a new fault occurs, the model is transferred to the new fault and updated using transfer learning and with a small database. To evaluate the performance of the proposed model, it was implemented and tested on the IEEE 39 bus system. The results show that the proposed model has high-performance accuracy and can be updated with a small database when encountering an unknown fault. According to the results, the proposed model has reduced the database size by 78.91% for optimal updating.

A Transfer Learning Method for Intelligent Load Shedding using Graph Convolutional Network considering Unknown Faults Keywords:

Graph Convolutional Network , Unknown Fault , Transfer learning , Event-based Load Shedding , Small Database