Fault location of flexible MTDC distribution network based on Bi-TCN-GRU

Increased distributed generation (DG) and storage deployment elevates distribution network requirements. Flexible DC distribution network has become the focus of contemporary research. However, the current fault diagnosis ideas for flexible DC distribution network are mostly used from AC system or DC transmission system. Therefore, a fault location method for multi-terminal flexible DC (MTDC) distribution network based on Bi-TCN-GRU is investigated. Firstly, the wavelet packet algorithm is utilised to decompose and reconstruct the line mode current, thereby yielding a series of reconstructed signals. Thereafter, the signal with the highest energy in the reconstructed signal is selected, and its power spectrum is calculated to obtain the feature vector. Secondly, the Temporal Convolutional Network (TCN) and the Gate Recurrent Unit (GRU) are combined to form a new network module, designated TCN-GRU. The TCN-GRU module adopts a bidirectional cyclic structure to construct a Bi-TCN-GRU network model. Finally, the Bi-TCN-GRU model is employed for the location of faults. The experiments show the method achieves high accuracy, strong fitting, and good noise immunity.