Abstract: To address the issue of insufficient frequency regulation capabilities and the need for more flexibility in existing frequency regulation resources within the new type of power system, there is an urgent need to tap into flexible frequency regulation resources. Large-scale electric vehicle (EV) clusters possess both the flexibility and rapidity required for frequency regulation resources, but the output of the EV cluster has strong uncertainty. Aiming at the above problems, the dynamic, outgoing model of EV under aggregator regulation is proposed to consider the two-stage optimal scheduling strategy for large-scale EV clusters to participate in frequency regulation service under user willingness. First, a user willingness classification prediction model is constructed based on user willingness data and support vector machine-random forest algorithm under snake optimization. Further, user willingness and frequency regulation capability are integrated to classify EVs into clusters, and a cluster dynamic frequency regulation output model is constructed with the actual charging strategy as a benchmark to evaluate the frequency regulation potential of clusters. Then, a day-ahead and intraday two-stage optimization scheduling model is proposed to minimize the fluctuation of remaining frequency regulation power demand. The global optimal scheduling plan is solved in the day-ahead stage; the intraday stage uses a rolling optimization method to locally optimize and correct the proposed under the existing frequency regulation market settlement method. Simulation examples have verified that the proposed scheduling strategy can improve the controllability of the EV cluster participating in frequency regulation services under the dispatch of the EV aggregator and has enhanced the economic benefits of both the EV aggregator and the EV cluster.