Research on Application of Large Model in Intelligent Fault Diagnosis of Power Equipment

Aiming at the complexity of power equipment fault diagnosis and the surge of data, this study proposes an improved large model "electro Transformer" based on transformer architecture, aiming at improving the accuracy and efficiency of power equipment fault diagnosis. In this study, the Transformer model is improved by introducing large model technology and combining the characteristics of power equipment fault signals. The ElectroTransformer model augments the encoding capability of multi-dimensional signals within the input embedding layer, enhances the self-attention mechanism through the introduction of multi-head attention and masking techniques, and employs a deeper network architecture coupled with residual connections and layer normalization to boost the model's expressive power and training stability. Experimental outcomes demonstrate that the ElectroTransformer outperforms traditional threshold-based diagnostic methods and support vector machines (SVM) in terms of accuracy, recall, and F1 score, achieving respective values of 0.925, 0.908, and 0.916. In addition, the effectiveness of the model in feature extraction is verified by dimensionality reduction visualization of t-SNE algorithm. The research shows that the large model technology has obvious advantages in intelligent fault diagnosis of power equipment, but it still needs to solve the challenges of data dependence, computing resource demand and interpretability in practical application.