Robust multi-time-scale dispatch of virtual power plants via entropy-guided rolling optimization and fuzzy-probabilistic control
Abstract
Aiming at the challenges of large day-ahead forecast errors, difficult multi-time-scale dispatch coordination, and state of charge (SOC) violation risks in energy storage equipped virtual power plants (VPPs), this paper proposes a multi-level robust optimal dispatch method based on rolling horizon optimization and model-driven fuzzy-probabilistic strategy real-time feedback (RHO-MFPS) for distributed energy systems scenarios. In the day-ahead planning stage, an optimization model is constructed to maximize the market revenue of the VPP. In the intra-day optimization stage, a two-layer dispatch framework is established based on the Markov decision process (MDP), integrating 30 min rolling optimization and 5 min ultra-short-term optimization to reduce modeling complexity. An entropy reward function is designed in the MDP model to characterize the rolling optimization process, which mitigates intra-day operation deviations induced by day-ahead forecast errors. Meanwhile, a model-driven fuzzy-probabilistic strategy (MFPS) is formulated for ultra-short-term optimization, which regulates the SOC of energy storage systems, balances charge–discharge capabilities, and enables accurate tracking of day-ahead dispatch plans. Simulation results demonstrate that the proposed RHO-MFPS method achieves a plan tracking deviation of 4.2%, an SOC violation rate of 1.8%, an average entropy reduction of 42%, and a total daily profit of 44 300 CNY. Under varying forecast error levels, the deviation increases by only 1.27% per 10% noise increment, verifying its superior robustness for distributed energy systems applications.