Fault Diagnosis Method for Voltage Transformers Based on Multi-modal Data Fusion and GoogLeNet

Voltage transformers are critical components in power systems, and their operational state directly impacts the safety and stability of the grid. Due to the potential occurrence of inter-turn short circuit faults during prolonged operation, fault diagnosis of voltage transformers has become a research focus to ensure reliable grid performance. This paper proposes a fault diagnosis method for voltage transformers based on multi-sensor data fusion and GoogLeNet. By collecting multi-modal signals such as voltage, current, and temperature, the method employs Continuous Wavelet Transform (CWT) to extract time-frequency features of the signals and utilizes the entropy weight method to weight and fuse the multi-modal features, thereby generating a comprehensive feature map that fully characterizes fault attributes. In constructing the diagnostic model, the pre-trained GoogLeNet deep neural network is fine-tuned to enhance classification performance on small-sample datasets. Experimental results show that the proposed method reliably diagnoses inter-turn short circuits at 25%, 50%, and 75% severity levels, achieving an average accuracy of 97.6%.