To address the issue of insufficient frequency regulation resources in power systems caused by high penetration of renewable energy, a two-stage optimization method is proposed based on virtual power plant （VPP） technology. This method aggregates flexible resources within buildings to jointly participate in multi-energy markets and frequency regulation ancillary service markets. Firstly, a gas-electric VPP model is constructed with smart buildings as units. Secondly, considering the high-quality frequency regulation potential of electric energy storage and electric vehicles within buildings, a day-ahead and real-time two-stage optimization strategy is designed for VPP-assisted grid frequency regulation. In the day-ahead stage, based on forecasts such as wind and solar generation, the mixed-integer linear programming method is applied to obtain the global optimal day-ahead scheduling scheme, aiming to minimize the total operating cost of the VPP. In the real-time stage, model predictive control combined with mixed-integer quadratic programming is used for rolling optimization based on real-time measurements such as frequency regulation signals, reducing prediction errors. Finally, simulation results under different scenarios demonstrate that the proposed method enables the VPP to fully exploit the flexibility potential of electric energy storage and electric vehicles for grid frequency regulation, achieving optimal resource allocation in both energy and frequency regulation markets while significantly improving the overall operational efficiency of the VPP.