Multi-modal fusion fault diagnosis for high-voltage transformers based on STFT-ResBIGRUNet

With the development of commercial complexes, higher demands are being placed on the stability and security of internal power systems. As a core component of distribution systems, high-voltage transformers play critical roles in energy conversion, power distribution, and fault isolation and protection. Meanwhile, the power consumption environments of critical users have become increasingly complex, characterized by diversified load types and frequently changing operating conditions, which present greater challenges to the reliable operation of high-voltage transformers. In this context, relying solely on a single signal source for fault diagnosis is no longer sufficient to meet practical requirements. Therefore, the introduction of multi-modal information fusion technology has become essential for improving diagnostic accuracy and comprehensiveness. To enhance the accuracy and intelligence of fault diagnosis in high-voltage transformers, this study proposes a multi-source information fusion-based fault diagnosis method based on the Short-Time Fourier Transform (STFT), Residual Network (ResNet18), and Bidirectional Gated Recurrent Unit (BiGRU), termed STFT-ResBiGRUNet. The proposed model utilizes fundamental electrical and environmental parameters, such as three-phase current, voltage, active power, reactive power, temperature, and humidity, as input features. By integrating the Efficient Channel Attention (ECA) mechanism, key feature representations are enhanced. Moreover, the model combines local feature extraction with the modeling of global temporal dependencies, thereby enabling efficient fault diagnosis of high-voltage transformers. Experimental results demonstrate that the proposed model exhibits significant robustness and high accuracy in noisy environments, outperforming state-of-the-art classification models.