In new power systems with a high proportion of power electronic interface devices, integrated photovoltaic-storage-charging systems face significant challenges in economic optimization due to their low inertia, weak damping physical characteristics, and highly nonlinear source-load fluctuations. Addressing the severe disconnect between long-term economic dispatch and short-term physical stability in existing systems, this paper proposes a hierarchical multi-scale collaborative dispatch architecture that integrates top-level Model Predictive Control (MPC) and bottom-level Grid-based Energy Storage (GFM) technology. In the top-level energy management domain, a robust MPC rolling optimization model considering battery dynamic degradation and multi-agent interaction between photovoltaic, storage, and charging is constructed. Through forward-looking power dispatch, a multi-objective Pareto equilibrium is achieved, considering grid purchase costs, battery life-cycle depreciation, and electric vehicle (EV) charging satisfaction. In the bottom-level physical execution domain, the discrete steady-state commands generated by MPC economic dispatch are innovatively seamlessly mapped to the dynamic active/reactive power reference dead zone of GFM control. This not only preserves hourly-level economic flexibility but also reshapes the system’s millisecond-level transient rigidity against voltage and frequency distortion. The full-dimensional multi-condition simulation results based on the real operation logs of the industrial park show that the proposed hierarchical architecture can effectively limit the frequency drop to within 0.18Hz under extreme off-grid switching conditions, and has both safety and robustness under different EV penetration rates. It also achieves a reduction of 18.4% and 12.5% in the average daily operating cost and battery aging expenses, respectively, providing a theoretical basis and engineering reference for the safe and efficient operation of large-scale photovoltaic energy storage charging and replenishment nodes.
