Transformer oil temperature prediction method based on DLinear and KAN

The transformer is a key component in the power system, and its operational state is directly related to the safety and stability of the power system. The top oil temperature is a key indicator for assessing the safe operation of the transformer. To predict the top oil temperature, a deep learning model combining DLinear and Kolmogorov-Arnold Network (KAN) is proposed. Firstly, the time series data are divided into trend and residual parts by the DLinear model, obtaining the linear trend and nonlinear residual after linear layer processing. Then, the residual data are processed by the KAN model. Afterward, through the dynamic weight fusion mechanism, the weights of the respective outputs of DLinear and KAN are dynamically adjusted to produce the prediction results. The results demonstrate that the oil temperature predictions generated by the DLinear-KAN model closely correspond to the actual values. In comparison with various other predictive models, the model proposed in this study exhibits superior prediction accuracy and minimal error, thereby substantiating its effectiveness.