Abstract: Accurate fault classification of intelligent electric meters has great significance to the accurate operation and maintenance of the metering equipment and the realization of the "Double Carbon" goals in China. The number of fault meter samples in different classes is extremely imbalanced, and the fault data presents the characteristics of multi-modal distribution and multi-class data overlapping in the hyperdimensional feature space, which brings great challenges to the accurate fault classification. A sample generation method based on the sample migration and boundary enhancement in overlapping areas is proposed. Different from the previous fault sample rebalancing methods that directly learn the distribution characteristics from the minority samples and generate the fault samples, this paper presents a sample overall balancing scheme that maps the majority samples to the boundary minority ones, which is more conducive to the classifier's learning of the class decision boundaries. Firstly, a domain-crossing consistency constraint for the latent vector of variational autoencoder (VAE) is designed, and a class-crossing sample generation network consisting of two pairs of VAEs sharing the latent vector space is constructed. In addition, the generative adversarial mechanism promotes the network to better learn the commonalities and differences between the samples in different classes. Based on this, a fault sample generation technique with the enhanced boundary of the data overlapping area is proposed. By introducing the Euclidean distance minimization constraint, the new minority samples are generated, and these samples form a clearer classification boundary from the original majority samples in the feature space. Experimental results on the 20 KEEL public imbalanced datasets and the fault datasets actually collected by the smart meters show that the proposed method outperforms the typical imbalanced classification methods.