<p>With the growing penetration of electric vehicles (EVs) into the power system, the complexity and diversity of EV charging behaviors present increasing challenges to grid stability. Understanding and analyzing these behaviors has therefore become essential. This study commences by constructing a charging behavior label system for EV users, employing preprocessed data collected from charging stations. We propose a clustering approach that integrates t-distributed stochastic neighbor embedding (t-SNE) for dimensionality reduction with affinity propagation (AP) clustering, referred to as the t-SNE-AP algorithm, to identify EV users with distinct charging characteristics. The influence of the t-SNE perplexity hyperparameter on dimensionality reduction is investigated, and AP clustering is applied to the reduced data. Dimensionality reduction evaluation indices are adopted to evaluate its performance. Clustering quality is assessed using the comprehensive evaluation index (CEI). After determining the optimal perplexity and clustering number, the t-SNE-AP algorithm is applied to the preprocessed charging data and benchmarked against other dimensionality reduction and clustering methods. The experimental results show that the proposed t-SNE algorithm achieves trustworthiness and continuity values of 0.998 and 0.987, respectively—the highest among all compared dimensionality reduction algorithms. Moreover, the CEI obtained by the proposed t-SNE-AP algorithm reaches 0.42, also demonstrating the best clustering performance compared with other clustering methods. Therefore, the proposed t-SNE-AP algorithm can effectively capture the intrinsic characteristics of EV charging behaviors and identify distinct clusters of EV users.</p>