Research on deep learning construction of energy user profiles considering environmental awareness

With the advancement of global dual-carbon goals, accurately characterizing the behavioral features and environmental preferences of energy users is crucial for demand-side management of smart grids. Existing energy user profiling methods usually rely on isolated user modeling and static behavioral descriptors, making it difficult to simultaneously capture the dynamic evolution of environmental awareness, spatial dependency among users, and long-term temporal behavior. These limitations reduce the accuracy and interpretability of user profiling in smart grid applications. To overcome these challenges, this study proposes a spatiotemporal deep learning framework (GCN-BiLSTM-Attention) that jointly models distribution network topology, temporal electricity consumption patterns, and dynamically updated environmental awareness features. It establishes a dynamic environmental awareness assessment and multi-dimensional user profiling system using a spatiotemporal deep learning framework (GCN-BiLSTM-Attention) that integrates graph convolutional networks and self-attention mechanisms. This successfully optimizes the joint extraction process of cross-node spatial spillover features and temporal features, significantly improving the accuracy of profile classification. Experimental results show that the model achieves an accuracy of 97.52% in user electricity behavior classification and profile clustering, with the silhouette coefficient increasing to 0.824. Compared to the traditional isolated node baseline model, the mean absolute percentage error (MAPE) is reduced by 5.82%. These findings provide precise guidance for regional-level green electricity collaborative marketing and demand response strategies based on distribution network topology, and also point to a new direction for user-side spatial behavior analysis in next-generation zero-carbon power systems based on graph computing.