A Few Shot Partial Discharge Diagnosis Method Based on Dual-Path Variational Autoencoder

Partial discharge (PD) diagnosis is critical for ensuring the reliability and safety of high-voltage equipment. However, the scarcity of labeled data, unseen fault classes, and cross-domain variability pose significant challenges to traditional diagnostic methods. This paper proposes a novel dual Variational Autoencoder (VAE) framework integrating a normal VAE for data augmentation and a Semi-supervised VAE for fault detection. The normal VAE generates synthetic samples to enhance dataset diversity, while Semi-supervised VAE leverages labeled and unlabeled data to improve generalization and adaptability to unseen faults. Experimental results demonstrate the superior performance of the proposed method in addressing data scarcity, enhancing generalization to unseen fault types, and achieving robust cross-domain diagnostics, establishing its potential for practical applications in PD diagnosis.