Application Research of Small-Sample Fault Diagnosis Method Based on Transformer-LSTM-SVM Hybrid Model

Conventional fault diagnosis techniques often depend on extensive datasets. However, acquiring fault data for critical or emerging equipment presents significant challenges. To tackle this problem, this study introduces a fault diagnosis model that integrates Transformer, LSTM, and SVM-referred to as the Transformer-LSTM-SVM method. This model integrates the Transformer's self-attention mechanism, the temporal modeling capabilities of LSTM, and the classification accuracy of SVM, aiming to improve diagnostic accuracy under small-sample conditions. Experimental results based on a publicly available PV power plant fault dataset demonstrate that the proposed method offers a clear advantage over other models in small-sample fault diagnosis. It significantly enhances fault recognition accuracy while improving the model's generalization ability and robustness. This study provides a novel approach to small-sample fault diagnosis in industrial applications and holds considerable engineering application potential.