To enhance the management efficiency of distributed energy resources, meet differentiated grid control requirements, and mitigate risks caused by wind-solar output uncertainties, a “resource-unit-platform” virtual power plant (VPP) energy management architecture is first proposed. The architecture enhances scheduling flexibility through interactions between the energy management platform and virtual unit aggregators. Second, a two-stage game strategy is designed. In the first stage, a master-slave (Stackelberg) game model is established between the platform and multiple virtual units. The platform adjusts the transaction prices offered to the units based on the power purchase and sale requirements reported by the aggregators and the dynamic electricity price of the distribution network. In the second stage, a generalized Nash bargaining cooperation game is introduced to transform the operational issues into one of cost minimization and benefit allocation. This reveals the interdependent effects of “distribution grid price fluctuations → platform price adjustment → unit transaction strategy adjustment”. Finally, by applying the (KKT) conditions and strong duality theory, the bi-layer optimization is reformulated as a single-layer mixed-integer linear programming problem, and conditional value-at-risk (CVaR) is incorporated to mitigate uncertainty-related risks. The proposed architecture and strategy form a coordinated mechanism of “external benchmark guidance → internal autonomous regulation → joint risk control”, offering significantly advantages in multi-resource co-optimization, risk management, and economic performance improvement.