制冷2026，Vol.45Issue(4)：22-28,7.DOI:10.3969/J.ISSN.1005-9180.2026.04.0005
基于STL分解及分项特征适配的冷负荷预测方法研究
Study on Cooing Load Forecasting Method Based on STL Decomposition and Sub-item Feature Adaptation
王宇凡 1丁云飞1
作者信息
- 1. 广州大学土木与交通工程学院,广东 广州,510006
- 折叠
摘要
Abstract
The energy consumption of Heating,Ventilation and Air Conditioning(HVAC)systems accounts for approximately 50～65％of the total building energy consumption.Accurate air conditioning load forecasting is essential for optimizing the operation of cold source systems.This study proposes a cold load prediction method based on STL decomposition and component-specific feature adaptation.The cooling load series is decomposed into trend,seasonal,and residual components using STL,and dedicated prediction models are assigned according to the physical characteristics of each component.An LSTM network is employed to capture the long-term smooth variations in the trend component.For the seasonal component,an LSTM model incorporating diurnal cycle time encoding and system operating condition encoding is used to characterize daily fluctuation patterns and equipment operating states.The residual component is modeled using a Random Forest that integrates date features and abnormal on/off indicators to capture random disturbances.The final prediction result is obtained by aggregating the outputs of all components.The results show that the component-specific characteristic modeling and feature introduction based on STL decomposition can significantly improve the prediction accuracy of cooling load.This model can provide support for complex time-series data prediction and the energy-saving operation of air conditioning cold sources.关键词
冷负荷预测/STL分解/长短期记忆网络/随机森林Key words
Cooling Load Forecasting/Stl Decomposition/Lstm/Random Forest分类
建筑与水利引用本文复制引用
王宇凡,丁云飞..基于STL分解及分项特征适配的冷负荷预测方法研究[J].制冷,2026,45(4):22-28,7.
