Multi-View Graph Attributed Learning Data Imputation Method for Residential Load Analysis

doi:10.1109/tsg.2026.3690142

Multi-View Graph Attributed Learning Data Imputation Method for Residential Load Analysis

2026-09-01

0

PRE

AI

DOI：10.1109/tsg.2026.3690142

Original

Original request for help

Save

Abstract

En 中文

With the proliferation of smart grids, residential load data has become vital for user behavior modeling and demand forecasting. However, real-world load curves frequently suffer from substantial missing values due to sensor or communication failures, making accurate imputation crucial. Existing data imputation methods encounter two main challenges. On the one hand, they often suffer significant performance degradation under high missing rates. On the other hand, they lack generalization across users, limiting their ability to accurately impute load curves for households with limited daily load curves. To overcome these challenges, this paper proposes a multi-view graph-attributed imputation framework. It constructs an observation-aware similarity graph using mask alignment to ensure structural reliability under sparse observations and introduces higher-order temporal gradient features to represent load dynamics. These views are fused through self-expressive learning and graph embedding to reveal latent user structures and guide robust imputation. Experiments on real-world residential data demonstrate the method's superior performance, particularly under severe missingness and limited target load data.

Keywords:

Filtering

Filters

Low-pass filters

Circuits and systems

Central Processing Unit

Protocols

Communication systems

Electronic mail

High frequency

Internet

Multi-view graph learning

residential load analysis

data imputation

Multi-View Graph Attributed Learning Data Imputation Method for Residential Load Analysis

Abstract

Journal

Organization

Cited Papers