Fault Identification Method for Flexible DC Grid Based on Convolutional Neural Network

Fast identification of DC faults has become a bottleneck restricting the rapid development of multi-terminal flexible DC grid. Based on the self-learning characteristics of convolutional neural network (CNN), a fault identification method based on convolutional neural network for flexible DC grid is proposed. The structure of the designed convolutional neural network consists of two input branches. The inputs of the two branches are a two-dimensional matrix composed of positive and negative line voltages and a two-dimensional matrix composed of positive and negative line currents. The sampling data window of voltage and current is set to 2 milliseconds after the start of the fault identification program. The scheme only utilizes single-ended electrical quantity, which not only realizes the rapid identification of DC fault types and areas, but also can reliably distinguish faults from lightning disturbance. Finally, a four-terminal flexible DC grid model is built in PSCAD/EMTDC to demonstrate the effectiveness of the fault identification method.