A Nonparametric Balanced Diffusion Based Fault Diagnosis Scheme for Electric Vehicle DC Charging Piles Under Imbalanced Samples

The imbalance between faulty and healthy samples of electric vehicle (EV) charging piles makes it difficult for both algorithm-level and data-level diagnostic methods to identify fault categories with limited data. In this article, a fault diagnosis scheme for EV DC charging piles based on a nonparametric balanced diffusion (BalDiff) model under imbalanced samples is proposed by the data-level method. First, the nonparametric BalDiff model is proposed to augment the imbalanced samples. Unlike existing generative data augmentation methods, it does not require the assumption of a specific data distribution. In addition, BalDiff uses attenuated noise instead of the constant noise intensity during the reverse process of the diffusion model to enhance the ability to capture the sample distribution. Second, an online fault diagnosis strategy applicable to different data sources is designed, which guarantees the diagnosis time while ensuring the diagnosis accuracy. Case studies on different data sources show that the proposed scheme achieves at least 4.2% and 6.7% improvement in F1-score and accuracy compared to the comparison methods, and the single diagnosis time is only 0.021 s.