Circuit breakers in converter station filter fields are critical protection and control components, and their operating condition directly affects the safety and stability of power transmission systems. In practical condition assessment tasks, field operational data are characterized by a severe class imbalance, where normal samples dominate and fault samples are extremely scarce, significantly degrading model accuracy and robustness. To address this challenge, an anomaly data augmentation approach based on a Conditional Wasserstein Generative Adversarial Network with Gradient Penalty (cWGAN-GP) is proposed. Multidimensional key features, including closing time and auxiliary switch operation time, are utilized. A small set of anomaly seed samples is first constructed using normal operation data combined with expert knowledge and manufacturer-defined thresholds. By introducing conditional constraints, the cWGAN-GP effectively learns the feature distribution of fault samples and generates high-quality synthetic data, thereby alleviating the imbalance problem. Furthermore, a Probabilistic Neural Network (PNN) optimized by the Dung Beetle Optimizer (DBO) is developed for circuit breaker health status identification. The DBO algorithm adaptively optimizes the key parameters of the PNN, improving classification performance. Experimental results demonstrate that the proposed framework achieves a macro F1-score of 88.21%, significantly outperforming benchmark models including Random Forest, ECOC-SVM, and XGBoost. The proposed method provides an effective solution to fault sample scarcity and offers a practical reference for intelligent condition assessment and operation and maintenance of power equipment.
Yu et al. (Wed,) studied this question.