Abstract: The aggregation schedulable capacity（ASC） of large-scale electric vehicles（EVs） is one of the important technical indicators for virtual power plant containing EVs to participate in multi-level and multi-scenario power balance auxiliary services. However,the existing ASC model of EVs is difficult to adapt to the interactive scenario between large-scale EVs and provincial power dispatching center. Therefore, from the new perspective of multi-level and multi-scenario regulation for peak-shaving, frequency regutalion, and voltage regulation of power systems, this paper proposes a dual-layer clustering modeling method for ASC of EVs based on data-driven and machine learning. By constructing the individual schedulable capacity model of generalized EV-charging pile energy storage unit, this method combines the density space-based clustering algorithm and the improved selforganizing map deep clustering algorithm, which effectively integrates the time distribution of electric quantity of EVs and the spatial distribution characteristics of charging piles, and constructs the ASC model for multi-scenario regulation of peak shaving,frequency regulation and voltage regulation. The proposed method is verified by actual charging records in a province of China, and various charging profiles such as “morning type”, “noon type” and “evening type” are obtained. The self-aggregation of generalized energy storage system with different spatial and temporal distributions of EVs is realized, and the potential evaluation of provincial-level EVs participating in different auxiliary services of power grid is realized. The data foundation is laid for the prediction of ASC.