With the advancement of high-penetration renewable energy grids, the technical feasibility and economic viability of electric vehicles (EVs) as flexible load resources participating in peak-shaving ancillary services have attracted significant attention. However, practical applications face three major challenges: (1) Coordination between real-time scheduling and day-ahead planning, where dynamic user behaviors lead to significant deviations between declared and actual peak-shaving capacities; (2) Conflict between privacy protection and regulation efficiency, as centralized control relies on user-sensitive data while decentralized strategies struggle to ensure global optimization; (3) Strict deviation penalties in peak-shaving markets, necessitating dynamic adaptation to mitigate risks. This paper proposes a hierarchical progressive strategy based on PID control, constructing a three-layer architecture of “day-ahead pre-scheduling, intra-day rolling optimization, and terminal execution”. The day-ahead layer integrates grid peak-shaving demands with time-of-use (TOU) electricity prices to generate initial charging/discharging plans that maximize revenue while constraining battery degradation and user incentive costs. The intra-day rolling optimization layer updates commands every 15 min, reducing computational complexity through K-means clustering-based EV cluster aggregation, dynamically allocates power weights via a rolling horizon model, and introduces PID feedback mechanisms to correct output deviations in real time, enhancing dynamic matching accuracy. The terminal execution layer coordinates charging stations and onboard battery management systems (BMS) to discretize continuous power commands into terminal charging/discharging rates, dynamically adjusts power limits based on SOC safety boundaries, and ensures user demand satisfaction and battery lifespan preservation. Simulation experiments based on Zhejiang Province grid data demonstrate: (1) The proposed strategy stabilizes peak-shaving output deviations within ± 15%, significantly outperforming traditional methods; (2) Incentive cost peaks decrease from 580 to 180 CNY, with single-cycle battery degradation reduced by 12.5%; (3) The hierarchical architecture achieves synergy between economic efficiency and grid security through cluster aggregation and localized decision-making without relying on user privacy data. The study validates the effectiveness of hierarchical PID control in dynamic peak-shaving capacity matching, deviation risk mitigation, and battery lifespan protection, providing technical support for refined virtual power plant (VPP) operations. Future work may extend to multi-timescale collaborative optimization and market mechanism design to further enhance the large-scale applicability of vehicle-grid interaction.