A Multiscale Diagnostic Approach for Oil-Paper Insulation Aging Using GAN-Enhanced Raman Spectroscopy

The aging of oil-paper insulation systems significantly influences the insulation performance of electrical equipment. Raman spectroscopy can detect changes in the composition of insulating oil, enabling the monitoring of the aging state of oil-paper insulation equipment. However, the scarcity of aging samples hinders the application of data-driven diagnostics. Therefore, this paper proposes a novel framework combining Wasserstein Generative Adversarial Networks with Gradient Penalty (WGAN-GP) and a multi-scale 1D Convolutional Neural Network (1D-CNN). The WGAN-GP generates high-fidelity synthetic samples, enlarging training datasets, and the multi-scale 1D-CNN is designed with parallel convolutional kernels of varying dilation rates (receptive fields of 3, 5, and 7) to capture both local peak features and global spectral trends. The proposed method achieved a diagnostic accuracy of 99.35% across ten aging stages. The diagnostic network was also evaluated using the training set and 20 real-world transformer samples, ultimately confirming the method’s ability to accurately diagnose the aging stages of oil-paper insulated electrical equipment.