Deep Learning-based Fault Detection, Classification, and Locating in Shipboard Power Systems

This paper proposes a new deep learning-based framework for fault detection, classification, and location identification simultaneously in shipboard power systems (SPS). Specifically, three different neural networks based fault detection methods, including deep neural network, gated recurrent unit, and long short-term memory, are developed and compared to detect different faults in SPS. The developed models use realtime line voltages of all SPS buses to detect faults for the entire network. The models are trained and tested based on simulated fault data from an 8-bus SPS. Results show that all the three methods have achieved a more than 99% accuracy in fault detection, classification, and location identification. The impacts of load variations and noisy inputs on the model performance are also investigated, and results show that the models are robust against the load variation and noisy input.