Fault Diagnosis of UHVDC Transmission System Based on Gated Recurrent Unit

The traditional method for detecting and classifying different types of faults in Ultra High Voltage Direct Current (UHVDC) has low accuracy rate and does not sufficiently exploit the information of timing characteristics in the electrical quantity signals, making it difficult to identify some system faults and measurement faults. This paper presents a fault diagnosis strategy which is based on Gated Recurrent Unit (GRU) to overcome these shortcomings. The study investigates three key GRU training factors, namely the amount of network layers, learning rate, and batch size, which may affect the training effect of GRU. And several sets of experiments are designed to determine the most suitable GRU network parameters. The classification effect of GRU is compared with that of RNN and LSTM, and the experimental results illustrate the high diagnostic accuracy of this approach. In addition, the method does not need to transform the data and can use the initial 1D time-series data as the network input, avoiding the loss of valid information and reducing the complexity of the fault diagnosis process.