Cooling load prediction model for multi-split systems in public buildings based on Transformer-xLSTM architecture

With the continuous application of multi-split systems in office buildings and other sectors, how to effectively utilize renewable energy under the background of its increasing integration into the energy structure has become a significant issue. This study, based on empirical data from a case building in Osaka, Japan, develops an hourly load prediction model that leverages the strengths of Transformer and xLSTM to effectively predict building cooling loads. The results show that the model achieves a coefficient of variation from root mean square error (RMSE-CV) of only 24.66%, demonstrating excellent predictive performance. This achievement not only provides a foundation for flexible regulation of the building, but also offers strong support for the application of multi-split systems in renewable energy integration and demand response.