煤气与热力2026，Vol.46Issue(6)：16-23,8.
热负荷预测模型输入变量选择及预测性能评价
Selection of Input Variables for Heat Load Forecasting Models and Evaluation of Forecasting Performance
摘要
Abstract
The input variables of building heat load fore-casting model were determined by correlation analysis,and the building heat load simulated by DeST software was used as sample data to establish a long-term and short-term memory(LSTM)network-based building heat load prediction model.The optimal hyperparam-eter combination of the LSTM network was selected.The input variables of the building heat load forecast-ing model include dry-bulb temperature,relative hu-midity,horizontal total radiation,occupancy status(oc-cupied or not),historical heat load at the 1st hour,his-torical heat load at the 2nd hour,and historical heat load at the 24th hour.The optimal hyperparameter combination is 16 nodes,learning rate of 0.001,batch size of 4,and 30 iterations.The LSTM network-based building heat load forecasting model demonstrates high prediction accuracy and excellent forecasting perfor-mance.关键词
建筑热负荷/相关性分析/长短期记忆网络/预测模型Key words
building heat load/correlation analysis/long short-term memory network/forecasting model分类
建筑与水利引用本文复制引用
王安庆,王海超,郎辉,李祥义..热负荷预测模型输入变量选择及预测性能评价[J].煤气与热力,2026,46(6):16-23,8.基金项目
科技部中芬政府间国际科技合作项目"基于数字孪生的供热系统全网动态优化及低碳智慧调控关键技术研究"(2021YFE0116200) （2021YFE0116200）
