@Article{ee.2026.083979,
AUTHOR = {Juan Li, Linghao Zhang, Anchi Ma, Li Ding, Yunlong Jiang, Jianyu Yu, Xinsong Zhang, Cheng Lu},
TITLE = {Adaptive Clustering of Distributed Energy Resources Fusing Temporal Behaviors and Geospatial Features},
JOURNAL = {Energy Engineering},
VOLUME = {},
YEAR = {},
NUMBER = {},
PAGES = {{pages}},
URL = {http://www.techscience.com/energy/online/detail/28376},
ISSN = {1546-0118},
ABSTRACT = {Against the backdrop of the global transition towards a low-carbon energy structure, the large-scale integration of Distributed Energy Resources (DERs) into the power grid introduces data characterized by high dimensionality, nonlinearity, and strong spatiotemporal coupling. These characteristics pose severe challenges to existing clustering methodologies. To address these challenges, this paper proposes an adaptive spatiotemporal clustering model for distributed resources that integrates temporal behavioral patterns with geospatial features. Firstly, the dataset undergoes preprocessing to satisfy the dimensional requirements of the model input. Subsequently, a ConvLSTM network is employed to perform deep feature extraction on the data. To further enhance feature quality, the Mamba state space model is introduced to adaptively denoise the feature sequences extracted by the ConvLSTM network and to capture global long-range temporal dependencies. The spatiotemporal feature vectors extracted by the ConvLSTM-Mamba architecture are then fused with vectors containing geospatial information via a weighted mechanism to generate the final feature representations for clustering. Furthermore, a contrastive learning strategy is adopted for model training. By maximizing the consistency between data-augmented samples, this approach drives the model to mine intrinsic structural features within unlabeled data, thereby fundamentally enhancing feature representation capability. Finally, utilizing the feature vectors extracted by the trained model, the OPTICS algorithm is applied to achieve adaptive clustering partitioning of the distributed resources. Case study analysis demonstrates that the proposed method significantly outperforms other state-of-the-art clustering techniques. The results indicate that the proposed model not only effectively enhances feature learning capabilities but also significantly improves the accuracy and robustness of the clustering results.},
DOI = {10.32604/ee.2026.083979}
}