Abstract: Due to the complex wiring order and method inside the transformer of the photovoltaic power station box, it is easy to confuse information and make mistakes during the diagnosis process, resulting in low accuracy of transformer winding fault diagnosis. Therefore, a fault diagnosis method of transformer winding based on double Kalman filter is proposed. The vibration signal of the transformer winding is processed by variational modal decomposition technology to extract modal functions and instantaneous frequencies, and a fault motion model of the transformer winding is established. The state of the transformer is predicted and corrected using the double Kalman filter algorithm, thereby achieving precise diagnosis of winding faults. The experimental results show that the designed method has a fault diagnosis accuracy of up to 98.57%, which is higher than the fault diagnosis accuracy of the other two diagnosis methods.