Fault Diagnostics in Shipboard Power Systems using Graph Neural Networks
Roshni Anna Jacob, Soroush Senemmar, Jie Zhang
Abstract
Shipboard power systems are evolving into sophisticated networks with automated protection and predictive control infrastructure. The need for real-time fault monitoring and detection in such systems can be facilitated by employing deep learning techniques. Taking into consideration the characteristic graph nature of the power network, this paper solves the fault detection and classification problem using graph convolutional neural networks. The proposed methodology translates the dynamic voltage measurements at the busbars of a shipboard power network along with the topology into input features for the learning framework. Both the type of fault and the location of the fault are determined. The developed model is validated on an 8-bus shipboard test network. The results indicate that the proposed algorithm has superior performance and can detect the fault type and location with an above 99% accuracy.