multi-terminal direct current grid based on modular multilevel converter （mmc-mtdc）, fault characteristics, empirical mode decomposition （emd）, kepler optimization algorithm （koa）, attention mechanism （am）, convolutional neural network （cnn）, bidirectional gated circulation unit （bigru）, fault diagnosis, Applications of electric power, TK4001-4102

Based on travelling wave features, a diagnostic method is proposed to address the complexity of manual threshold setting process and the difficulty of detecting high-resistance faults in the fault diagnosis of multi-terminal direct current grid based on modular multilevel converter （MMC-MTDC）. Firstly, the blocking effect of boundary elements on high-frequency signals is identified by analyzing the fault characteristics of the system. Secondly, empirical mode decomposition （EMD） is employed to decompose power signals into intrinsic mode function （IMF）, and the energy values of the IMF is utilized as fault features to train the CNN-BiGRU network composed of convolutional neural network （CNN） and bidirectional gated recurrent unit （BiGRU）. On this basis, the Kepler optimization algorithm （KOA） and attention mechanism （AM） are employed to enhance the CNN-BiGRU network to realize the fault diagnosis of the MMC-MTDC. Finally, the simulation model is built in PSCAD/EMTDC. The results show that the method can not only realize the detection of bus faults and line faults but also solve the problem of easy refusal of protection under the high resistance state while meeting the requirements of protection reliability and speed.