Transformer vibration modeling and condition assessment using residual neural networks

Transformer vibration signals can reflect various internal faults but are also influenced by non-fault factors. To improve the accuracy of transformer condition assessment, it is necessary to propose a method that accounts for non-fault factors. This paper first establishes a mathematical model of transformer vibration, and analyzes the relationships between influencing factors (including both fault-related and non-fault-related factors) and vibration signals. A vibration model incorporating non-fault factors is developed using a Residual Neural Network (ResNet). The model simulates normal transformer vibrations, enabling condition assessment by comparing simulated signals with measured data. Experimental results demonstrate that the proposed vibration model accurately simulates normal transformer vibration signals, and the assessment method effectively distinguishes between different operational states of transformers.