Fault Diagnosis Method for High-Voltage Circuit Breakers Based on Physics-Informed Transfer Learning

Dong Wang^*Lubo ZhouLiyun XieXipu LiuShiqi DongJunhua Liu

• State Grid Shanghai Ultra High Voltage Company, Shanghai, China

The running structure of high-voltage circuit breakers is an important part in power systems, and its machine condition directly affects the device steadiness and the secure work of power grids. To raise the exactness of trouble checking for high-voltage circuit breakers, this paper suggests a trouble checking way based on many-source information and movement study. First, voiceprint and current signals are analyzed using a single-dimension rolling nerve network (1DRNN) to extract feature data related to the mechanical condition of the operating mechanism. Second, combined with mechanical parameter identification, a physics-informed transfer learning network model is proposed, which consists of a Common Feature Learning Network (CFLN) and a Mechanical Feature Learning Network (MFLN). The model is designed to learn shared features between multi-source signals and mechanical parameters, while the MFLN focuses on extracting specific features of individual mechanical parameters. Additionally, a multi-head attention mechanism is integrated to enhance the model's ability to capture key features, and a physics-based loss function is adopted to improve the physical consistency of the model during mechanical parameter identification. Experimental results demonstrate that the proposed method achieves a fault diagnosis accuracy of over 93% and maintains high stability and accuracy even under noise interference.