Voltage Sag Recognition Based on CM-eCNN Model For Few-Shot Learning

Abstract

Voltage sag is a serious problem in power systems. To reduce voltage sag, it is necessary to identify and classify voltage sag sources. However, due to insufficient voltage sag data, our paper designs a CM-eCNN model through few-shot learning. Firstly, this paper introduces the CM-eCNN model and studies how to improve the extraction efficiency of the constellation model to feature map through the attention mechanism. Then, it introduces the training method of few-shot learning. Finally, our model training with few-shot learning has better performance than other models, which proves effectiveness of the model.