用于建筑能耗预测的多尺度可解释时序预测网络模型

打开文本图片集

中图分类号：TM715 文献标志码：A 文章编号：1000-582X（2026）04-026-11

doi：10.11835/j.issn.1000-582X.2026.04.003

Multi-scale interpretable temporal prediction network for building energy consumption forecasting

YANG Liejuan ，TAN Guopeng'，CAO Qi'，YANG Huiyue'，ZHOU Yang

（1.JointLogistic SupportForce UniversityofEnginering，Chongqing 401331，P.R.China;2.Unit 78156of the

Chinese People’sLiberation Army，Chongqing 40o039，P.R.China; 3.Chongqing Construction Science Research Institute Co.，Ltd.， Chongqing Design Group Co. Ltd.， Chongqing 400042，P.R.China）

Abstract：Accurate forecasting of building energy consumption is crucial for optimizing energy management， reducing operational costs，and achieving carbon neutrality goals.This study proposes a multi-scale interpretable temporal prediction network model （ITSFN），which enhances prediction accuracy and reliability through the collaborative optimizationof long short-term temporal （LSTM） networksand Kolmogorov-Arnold networks （KAN）.The model integrates temporal-environmental feature decoupling with a dynamic attention mechanism， explicitly decomposing time-series data into seasonal，trend，andresidual components to construct a structured feature space.It employs a parallel architecture of gated recurrent units （GRU）and multi-head atention to model multi-scale features.Tested on an energy consumption dataset from a university building in a hot-summer/coldwinter region，ITSFN outperforms traditional models： it reduces the root mean square error （RMSE） of total energy consumption prediction by 13.9% compared to LSTM and decreases the RMSE of sub-item energy consumption prediction by 31.1% compared to Transformer. Additionally， ITSFN enhances the noise suppression coefficient to0.89 through feature decoupling，achieves a local attention angle of 0.92 in mutation regions，and reduces over-smoothing by 29.6% compared to traditional methods.By quantifying feature contributions， the model reveals theevolutionary patterns of component weights，further validating its efctivenessand practical applicability.

Keywords： interpretable temporal prediction; feature decoupling; hybrid attention mechanism; long short-term memory （LSTM）; Kolmogorov-Arnold network （KAN）

建筑能耗是中国能源消耗总量的主要来源之一，主要有空调系统、照明系统、运维设备与其他设备等消耗形式。（剩余11986字）