[1]赵 冬,李亚瑞,王文相,等.基于动态融合注意力机制的电力负荷缺失数据填充模型[J].郑州大学学报(工学版),2024,45(pre):1-8.[doi:10.13705/j.issn.1671-6833.2024.05.004]
　ZHAO Dong,LI Yarui,WANG Wenxiang,et al.Dynamic Fusion of Attention Mechanism Imputation Model forMissing Data of Power Load[J].Journal of Zhengzhou University (Engineering Science),2024,45(pre):1-8.[doi:10.13705/j.issn.1671-6833.2024.05.004]
点击复制

基于动态融合注意力机制的电力负荷缺失数据填充模型()
分享到：

《郑州大学学报(工学版)》[ISSN:1671-6833/CN:41-1339/T]

卷:
45
期数:
2024年pre
页码:
1-8
栏目:
出版日期:
2024-12-31

文章信息/Info

Title:
Dynamic Fusion of Attention Mechanism Imputation Model forMissing Data of Power Load
作者:
赵 冬;  李亚瑞;  王文相;  宋 伟
1. 中原工学院 软件学院2. 中原工学院 计算机学院3. 许昌许继软件技术有限公司4. 郑州大学 计算机与人工智能学院
Author(s):
ZHAO Dong;  LI Yarui;  WANG Wenxiang;  SONG Wei
1. School of Software, Zhongyuan University of Technology, Zhengzhou 450007, China; 2. School of Computer, Zhongyuan Universityof Technology, Zhengzhou 451191, China; 3. Xuchang Xuji Software Technology Co. , Ltd. , Xuchang 461000; 4. School of Computerand Artificial Intelligence, Zhengzhou University, Zhengzhou 450001, China
关键词:
缺失值填充 注意力机制 电力负荷 时序特征
Keywords:
Missing value filling;  Mechanisms of attention;  Electrical load;  Temporal characteristics
DOI:
10.13705/j.issn.1671-6833.2024.05.004
文献标志码:
A
摘要:
为了提高电力负荷数据的缺失值填充精度,保障后续数据分析与应用的高效进行,提出一种基于动态融合注意力机制的填充模型 DFAIM。 该模型由注意力机制模块和动态加权融合模块构成,通过注意力机制模块的2 种不同注意力机制挖掘特征与时间戳之间的深层关联通过动态加权融合模块将可学习的权重赋予注意力机制模块的 2 个输出以得到特征表示最后,利用特征表示来替换缺失位置的值,从而得到准确的填充结果。 使用纽约市某地区的气象及负荷数据集及 UCI 电力负荷数据集对提出的模型进行验证,实验结果表明:相较于统计学、机器学习和深度学习填充模型,DFAIM 在评价指标 MAE、RMSE 和 MRE 上均具有一定优势。
Abstract:
In order to improve the accuracy of missing value filling of power load data and ensure the efficient follow-up data analysis and application, a filling model based on dynamic fusion attention mechanism is proposed. The model consists of an attention mechanism module and a dynamic weighted fusion module, and the deep association between features and timestamps is mined through two different attention mechanisms of the attention mechanism module. The learnable weights are assigned to the two outputs of the attention mechanism module by the dynamic weighted fusion module to get the feature representation. Finally, the feature representation is used to replace the values at the missing positions to obtain accurate filling results. The proposed model is validated using the meteorological and load dataset of a certain area of New York City and the UCI power load dataset, and the experimental results show that DFAIM has certain advantages over statistical, machine learning, and deep learning filling models in MAE, RMSE, and MRE.
更新日期/Last Update: 2024-05-23