华中科技大学学报（自然科学版）2026，Vol.54Issue(1)：1-7,7.DOI:10.13245/j.hust.250141
考虑数据时空分布特征的建筑群电力负荷填补方法
Filling method for electric load of building clusters considering spatiotemporal distribution characteristics
苏丽弘 1刚文杰 1徐新华 2张颖 3董书琨2
作者信息
- 1. 华中科技大学人工智能研究院,湖北 武汉 430074
- 2. 华中科技大学环境科学与工程学院,湖北 武汉 430074
- 3. 华中科技大学长江流域多介质污染协同控制湖北省重点实验室,湖北 武汉 430074
- 折叠
摘要
Abstract
To assist power companies and load aggregators in addressing the issue of missing data in building clusters,a filling method,which utilized the spatiotemporal distribution characteristics of the data,was proposed for imputing missing power load data.Based on the spatiotemporal features of the data,the building clusters load data were divided for learning and reconstruction.Based on K-SVD(K-singular value decomposition),a load dictionary was constructed by learning from complete samples to capture power consumption patterns.The observed portions of incomplete samples were utilized to identify consumption patterns,which can facilitate the reconstruction of the missing data.The effectiveness of the proposed method was validated by real-world power load data collected from a specific region,and its imputation performance was compared and analysed in both spatial and temporal dimensions.The results demonstrated that the proposed K-SVD-based imputation method achieved low imputation errors and maintained low computational costs.When the missing rate was 70%,the MAPE and CVRMSE of the temporal K-SVD reached 3.7%and 4.5%,respectively.The spatial K-SVD performed better than the temporal K-SVD,with maximum reductions in MAPE and CVRMSE of up to 9%and 10.1%,respectively.关键词
数据填补/K-奇异值分解/电力负荷/建筑群/时空分布特征Key words
filling data/K-singular value decomposition(K-SVD)/electric load/building clusters/spatiotemporal distribution characteristics分类
信息技术与安全科学引用本文复制引用
苏丽弘,刚文杰,徐新华,张颖,董书琨..考虑数据时空分布特征的建筑群电力负荷填补方法[J].华中科技大学学报（自然科学版）,2026,54(1):1-7,7.
