A Novel Power Equipment Heating Trend Prediction Model Based on Improved Informer Algorithm

Power equipment is an important part of the power system, and its impending failures are often accompanied by abnormal heating. Therefore, predicting and tracking the future heating trends of power equipment is important for the stable operation of the power system. In this paper, an improved Informer model, Informer-KDC, is proposed for heating trend prediction of power equipment. In Informer-KDC, first, dilated causal convolution is used to increase the receptive field of the model. Secondly, a new residual connection is introduced to preserve more original feature information and weaken the possibility of network degradation. Finally, the Kernel-MSE loss function is utilized to address the nonlinearity of errors, enhancing the model's predictive performance. The proposed Informer-KDC model improves upon the Informer model in both network architecture and loss function, significantly increasing prediction accuracy. Experiments conducted on the widely used ETT long-term time series dataset and the ECL power load dataset demonstrate that the Informer-KDC model exhibits outstanding predictive performance, achieving the best performance metrics on both datasets.