Transformer Fault Identification Method Based on KPCA-IDBO-KELM

To improve the fault diagnosis accuracy of oil-immersed transformers, a transformer fault diagnosis method based on KPCA-IDBO-KELM is proposed. Firstly, the high-dimensional nonlinear fault features are downscaled by using kernel principal component analysis (KPCA); secondly, the dung beetle optimization algorithm is improved by introducing random wandering and longitudinal and transverse crossover strategies, and the improved IDBO is compared with the original algorithm and the other two algorithms. Finally, the feature data processed by KPCA are used as inputs, and the kernel function parameters and regularization coefficients of KELM are optimized by IDBO to establish the KPCA-IDBO-KELM transformer fault identification model. The experimental results show that the accuracy of KPCA-IDBO-KELM diagnostic method reaches 95.0%, which can more accurately identify the transformer fault state compared with other diagnostic models.