A Hierarchical Charging Station Identification Method Integrating Charging-Order Behaviour Mining and Deep Multi-View Clustering

Ultra-fast electric vehicle (EV) charging can create spatially concentrated high-power loads and increase stress on distribution grids. However, charger rated power alone cannot adequately represent the actual operation of a charging station. This study proposes an interpretable order–station two-level clustering framework for identifying charging-station types from large-scale operational data. At the order level, Balanced Iterative Reducing and Clustering Using Hierarchies (BIRCH) identifies charging patterns using average actual charging power, delivered energy, and power utilization. The resulting order clusters are aggregated by station to characterize the internal composition of charging demand. At the station level, a deep multi-view clustering model integrates four complementary views: charging-order composition, actual operation, equipment capability, and temporal operation. The temporal-operation view is extracted from charging-load sequences using a long short-term memory (LSTM) autoencoder. The framework is evaluated through cluster-quality metrics, distribution-shift analysis, characteristic-range checks, and business-consistency constraints. The results indicate that combining order-level behavioural information with complementary station views improves cluster separation and stability relative to conventional clustering methods. Finally, the clustering labels are distilled into fixed decision-tree rules, providing explicit and interpretable thresholds for identifying new stations. Furthermore, the identified station classes exhibit distinct electrical impacts in terms of transformer loading, feeder loading, voltage deviations, and network losses. Under the adopted benchmark conditions, class-informed capacity sizing reduces aggregate transformer capacity and feeder ampacity requirements by 25.10% compared with uniform worst-case sizing, demonstrating the potential value of the proposed classification for differentiated distribution-grid capacity assessment.