User behavior profiling and demand response potential evaluation for electric vehicles in new-type power systems: A SOM–GMM and entropy-based approach

With the rapid growth of electric vehicles (EVs), large-scale and uncoordinated charging has intensified grid peak-load pressure, highlighting the need for effective demand response (DR) strategies. However, existing studies often oversimplify user behavior and fail to capture the heterogeneity of charging flexibility. To address this gap, this paper proposes a data-driven framework for evaluating EV users' DR potential from both time flexibility and power scalability dimensions. First, a comprehensive indicator system is constructed to quantify charging behavior features across temporal and power domains. Then, a self-organizing map–Gaussian mixture model (SOM–GMM) hybrid clustering method is employed to classify users, yielding eight representative user clusters with distinct charging preferences and flexibility characteristics. Subsequently, an information-gain and redundancy-based feature optimization procedure is implemented to select six key behavioral indicators, ensuring high interpretability and minimal redundancy. Finally, the entropy-weighted TOPSIS approach is applied to comprehensively evaluate each cluster's DR potential. The results show that clusters with long charging durations and high adjustment rates exhibit the highest effective DR potential, while rigid peak-charging users have limited time-flexible response capability. The proposed method not only improves the precision and interpretability of user classification but also provides a quantitative basis for differentiated DR pricing, incentive design, and flexible load management. This study offers valuable insights for enhancing user-side flexibility utilization and supports the construction of new-type power systems integrating distributed and flexible EV resources.