制冷技术2025，Vol.45Issue(4)：39-45,7.DOI:10.3969/j.issn.2095-4468.2025.04.201
基于序列分解和趋势混合的建筑能耗长期预测研究
Research on Long-Term Prediction of Building Energy Consumption Based on Sequence Decomposition and Trend Fusion
摘要
Abstract
A model based on sequence decomposition and trend fusion is proposed to enhance the accuracy of long-term building energy consumption forecasting.The energy consumption data undergoes dual encoding of features and time,followed by parallel processing through sequence decomposition and fusion computation pathways.The outputs of both pathways are integrated to yield the prediction results.Comparative experiments demonstrate that the proposed model achieves a maximum reduction of 25.7%in mean absolute error(MAE),40.2%in mean squared error(MSE),and a maximum improvement of 36.9%in the coefficient of determination(R²).To further validate the rationality of the model's architecture,ablation studies are conducted.The results indicate that,compared to models without temporal encoding,the proposed model R² is improved by 32.7%,MAE is reduced by 27.9%,and MSE is decreased by 50.9%.Additionally,compared to models without sequence decomposition and trend hybridization modules,a further enhancement of 9.7%in R²,a reduction of 5.2%in MAE,and a decrease of 20%in MSE are achieved by the proposed model.关键词
建筑能耗/序列分解/长期预测Key words
Building energy consumption/Sequence decomposition/Long-term energy forecasting分类
能源科技引用本文复制引用
张哲铭,刘颖,王文彬,晋欣桥,杜志敏..基于序列分解和趋势混合的建筑能耗长期预测研究[J].制冷技术,2025,45(4):39-45,7.基金项目
科技部重点研发计划(No.2021YFE0107400),国家自然科学基金(No.52276011). （No.2021YFE0107400）
