Addressing the scarcity of transformer fault image samples, this technology proposes an innovative adversarial network method. By introducing SimAM attention mechanisms in the generator, combining adaptive data augmentation strategies and LeCam regularization loss functions, it effectively improves image generation quality and speed under small-sample conditions. The generated fault images are highly realistic and can be used to build large-scale diagnostic databases, overcoming the sample bottleneck for AI in transformer fault diagnosis and providing critical technical support for power system safety.

Technology provider:Nanjing University of Posts and Telecommunications