An Improved CWT-ViT Method with Residual Attention for Synchronous Generator Fault Diagnosis

Under high load and complex operating conditions, in order to meet the smart grid's demand for high reliability and automation, and ensure the continuity and stability of power supply, it is crucial to perform fault diagnosis of synchronous generators. This paper addresses the issue of insufficient capability in extracting low-level information from synchronous generator fault data using traditional Vision Transformer (ViT) models, and proposes an improved ViT model. The method first transforms the raw signal into a time-frequency image using Continuous Wavelet Transform (CWT), and then optimizes the attention mechanism in ViT by introducing residual connections to enhance the model's ability to focus on the low-level information of the data. Experimental results show that the improved model achieves an accuracy of 96.38% in synchronous generator fault diagnosis, demonstrating significant advantages over traditional methods and proving its excellent performance in fault diagnosis of synchronous generators.