A novel heat load prediction algorithm based on fuzzy C-mean clustering and mixed positional encoding informer

With the rapid urbanization in China, sustainable energy development and environmental challenges arising from the massive amount of energy consumed by district heating systems (DHS) have attracted widespread attention. An accurate heat load prediction algorithm contributes to the optimized regulation of DHS. However, heat load has significant nonlinear and thermally inertia properties, and traditional prediction models suffer from weak generalization ability and low sensitivity to dynamic changes. In this paper, a hybrid heat load prediction model based on fuzzy C-mean clustering and mixed positional encoding Informer (FCM-MPE-Informer) is innovatively proposed by integrating the advantages of supervised and unsupervised learning. Two different mixed positional encoding strategies were first designed to improve the Informer and investigated in detail to explore the enhancement of their sensitivity to changes in long distance series. The model sufficiently incorporates the strengths of the FCM algorithm in dealing with uncertainty and complex relationships while incorporating the mixed positional encoding Informer algorithm to efficiently handle the interdependencies between the long time series of heat load and influencing factors. The comprehensive prediction performance comparison and ablation experimental results sufficiently validate the superiority of the proposed FCM-MPE-Informer prediction algorithm. The mean absolute percentage error (MAPE) indicators for the four heat exchange stations were 2.575 %, 2.510 %, 2.584 %, and 2.640 %, respectively.