Abstract: The lack of electricity theft data has a great impact on the identification accuracy of the electricity theft detection algorithm. Therefore, this paper proposed an electricity theft detection method based on the triplet network for few-shot learning. Firstly, the gramian angular field was used to convert the electricity consumption sequence to images. Secondly, the triplet network was used to extract the feature vector of the user's electricity consumption data and measure their similarity based on the Euclidean distance. Finally, electricity theft detection was realized. The triplet network not only extracts the feature of the training samples, but also learns the similarity between the samples of the same class and the difference between the samples of different classes, thus improving the clustering effect of the feature vectors and showing a higher silhouette score. The results of calculation examples verify the accuracy and superiority of the proposed algorithm for few-shot learning.