Considering Photovoltaic Uncertainty in Two-stage Optimal Scheduling of Virtual Power Plants

The uncertainty of photovoltaic (PV) output severely constrains the economic dispatch and stable operation of virtual power plants (VPPs). Existing research on modeling renewable energy uncertainty suffers from limitations such as strong data dependency and insufficient multi-period coordination. Moreover, the complex spatiotemporal constraints of electric vehicle (EV) resources make them difficult to aggregate efficiently. This paper proposes a two-layer optimal scheduling framework that integrates PV uncertainty modeling and dynamic EV aggregation. For PV output uncertainty analysis, a hybrid VMD-SSA-LSTM model is constructed. Variational mode decomposition (VMD) and singular spectrum analysis (SSA) are employed to process multi-dimensional time-series data, extracting intrinsic mode features that are then fed into an LSTM network to accurately capture the spatiotemporal correlations of PV output. Additionally, a dynamic schedulable potential model is established based on Minkowski summation, aggregating dispersed EVs into a generalized energy storage unit. Case study results demonstrate that the proposed model, through a synergistic mechanism of high-precision PV forecasting, dynamic EV aggregation, and carbon cost constraints, provides an economic, stable, and low-carbon scheduling method for VPPs with high penetration of renewable energy. The effectiveness of the model in enhancing both economic efficiency and low-carbon performance is verified.