Abstract: With the growing integration of renewable energy and the expanding scale of new power loads, the peak-valley difference of the net load in power systems has widened, posing severe challenges to the power grid's secure and economic operation. This paper proposes a power-sharing dispatch model for multi-park-level virtual power plants (VPPs) aimed at distribution network peak management. First, a localized dispatch model for individual park-level VPPs is constructed, where mixed-integer linear constraints are used to extract net load peak variables and achieve localized peak management. Then, a power-sharing mechanism among VPPs is introduced to extend the localized model into a centralized optimization dispatch model for multiple VPPs, which is associated with energy storage system (ESS) charging/discharging to effectively shave peak loads. Finally, based on a case study of park-level VPP systems with photovoltaic (PV) generation, energy storage, and loads, the benchmark, localized dispatch, and power-sharing dispatch models are compared. Simulation results show that the power-sharing scheme reduces total operating costs by 17.8% and 5.89% relative to the benchmark and localized models, respectively; and the daily peak demand of the distribution network decreases by 5.97% compared to localized dispatch. While reducing the total operating costs of multiple VPPs, the proposed method effectively manages the distribution network peak loads.