To address the issues of slow processing speed and low accuracy when a large number of electric vehicles (EVs) are integrated into the power grid under vehicle-to-grid (V2G) scenarios, a dynamic EV classification and multi-step Markov chain aggregation method based on a density-based spatial clustering of applications with noise (DBSCAN) algorithm is proposed. In the classification phase, the DBSCAN algorithm is improved using the k-distance curve and its differential form, and the concept of incremental clustering is introduced to dynamically classify EV data, resulting in EV clusters characterized by multi-dimensional features such as state of charge (SOC), remaining connection time, and controllable capacity. In the aggregation phase, a multi-step state transition Markov chain theory is developed to construct online aggregation models for each EV cluster. This approach addresses the limitations of traditional Markov chains in handling multi-feature EV aggregation and improves the accuracy of the aggregated power output. Simulation results demonstrate that the proposed classification method can quickly and accurately partition large-scale EVs integrated into the grid into different clusters, and that the aggregation model significantly improves the accuracy of aggregate power estimation, effectively addressing the challenges associated with large-scale EV integration.