MBDiff: A Multi-View Behavior-Aware Diffusion Model for Utility Data Imputation

A new research paper on arXiv (2607.29177) introduces MBDiff, a multi-view behavior-aware diffusion model designed for probabilistic imputation of utility data, such as electricity, water, and gas consumption. The study addresses the common problem of missing values in data collected by sensors and embedded devices, which can arise from device failures or transmission issues. These gaps can negatively impact billing accuracy, demand forecasting, and supply management. While existing imputation methods often rely on aggregated datasets, MBDiff incorporates rich user behavior information to improve accuracy. The authors highlight the challenge of learning comprehensive user behavior from long-term, diverse, and incomplete data, and note that leveraging such behavior to guide imputation is nontrivial due to its indirect nature. The paper is categorized as a cross-type announcement and is available on arXiv. The model's approach is probabilistic, suggesting it provides uncertainty estimates for imputed values. The research is relevant to both industry and academia, as utility data imputation is a topic of significant interest. The paper does not specify a publication date or authors in the provided content.