Flexible DC distribution network fault detection method based on MTF-EfficientNetV2 algorithm

Abstract

Given the swift advancement of clean energy, flexible DC distribution network has become a research hotspot for future power grids. Existing DC line fault detection methods have problems such as low detection precision and vulnerability to resistance. For this reason, a fault detection method built on the upgraded EfficientNetV2 algorithm is proposed. Primarily, the fault transient voltage time-domain data are gathered. To enhance the variability of fault features, the data are transformed to a two-dimensional image by Markov variation field. Then, a dual-channel attention mechanism is used to shortlist and fuse the features with channel and spatial features, respectively. Finally, the fused features are fed into EfficientNetV2 for training. And the detection results are obtained by testing the model under different working conditions. The findings demonstrate the excellent detection accuracy of the approach. The average accuracy can reach 98.95%.