Power Load Missing Data Imputation Model Based on Dynamic Fusion Attention Mechanism

• Published in: Zhengzhou Daxue Xuebao. Gongxue Ban = Journal of Zhengzhou University. Engineering Science Vol. 46; no. 2; p. 111
• Main Authors: Zhao, Dong, Li, Yarui, Wang, Wenxiang, Song, Wei
• Format: Journal Article
• Language: Chinese; English
• Published: Zhengzhou Zhengzhou University 01.01.2025
• Subjects: Data analysis; Datasets; Deep learning; Electrical loads; Machine learning; Missing data; Modules; Representations
• ISSN: 1671-6833
• DOI: 10.13705/j.issn.1671-6833.2024.05.004

In order to improve the imputation accuracy of power load missing value data and guarantee the efficiency of subsequent data analysis and application, firstly, a imputation model based on dynamic fusion of attention mechanism dynamic fusion of attention mechanism imputation model (DFAIM) was proposed. The model consisted of an attention mechanism module and a dynamic weighted fusion module, where the deep correlation between features and timestamps was mined through the two different attention mechanisms of the attention mechanism module. Secondly, feature representations were obtained by assigning learnable weights to the two outputs of the attention mechanism module through the dynamic weighted fusion module. Finally, replacing the values of the missing locations with the obtained feature representations to obtain the imputed values. The proposed model was validated using the meteorological and load dataset and the UCI electric load dataset for an area in New York City, and the experimental results showed that DFAIM had certain advantages in evaluating metrics such as MAE, RMSE, and MRE compared to statistics, machine learning, and deep learning models imputation models.