ABSTRACT With the advancement of smart grid and Internet of Things, alongside broad adoption of distributed energy resources, precise profiling of residential users has become vital to grid operational efficiency and load forecasting accuracy. However, existing profiling approaches mainly rely on single‐source load data and fail to capture the dynamic behavioural heterogeneity of modern prosumers, limiting forecasting performance and demand‐side management effectiveness. This article proposes a data‐driven multidimensional profiling framework integrating energy configuration attributes, multisource heterogeneous data validated by Analysis of Variance and dynamic consumption patterns identified through two‐stage clustering algorithms combining Dynamic Time Warping Path‐Aware Weighted K‐Shape with Agglomerative Hierarchical Clustering. Robust data preprocessing is ensured via an Improved Quantum‐Inspired Genetic Algorithm coupled with Random Forest Regression and Empirical Mode Decomposition. Case studies using an Australian dataset demonstrate that integrating the generated user profile labels into a Transformer‐based forecasting model reduces normalised root mean square error by 41.6% compared to baselines. The framework successfully characterises complex user behaviours, providing a scalable methodology for load forecasting, personalised energy services and precise demand response while enabling utilities to better accommodate distributed resources and support renewable energy integration critical to power system stabilisation.
Li et al. (Mon,) studied this question.