Abstract
The variability of renewable energy sources presents major challenges for accurately evaluating the maximum dispatchable capacity of the Virtual Power Plant (VPP). This study proposes a scenario-driven framework to assess the maximum dispatchable capacity of a VPP under combined wind, solar, gas, and storage. First, a hybrid deep learning model combining Adaptive Graph Convolutional Networks (AGCN), Convolutional Neural Networks (CNN), and Long Short-Term Memory (LSTM) is used to forecast short-term wind speed and solar irradiance. Load uncertainty is modeled by applying random perturbations to a typical winter daily profile. Second, the forecast results are used to construct a probabilistic ensemble of source-load trajectories, which is statistically reduced to a compact set of representative scenarios while preserving the key temporal variability of renewable generation and its correlation with demand. A multi-scenario stochastic optimization model is then used to evaluate dispatch feasibility across all reduced scenarios. Finally, a binary search strategy combined with feasibility screening is employed to determine the maximum dispatchable capacity that can be reliably committed across all uncertainty conditions. Simulation results confirm the effectiveness of the proposed method in supporting reliable, economically feasible capacity planning for the VPP under uncertainty.