A proactive real-time dispatch approach with rapid flexibility projection of energy storage systems

High renewable penetration increases intra-day net-load volatility and forecast errors, making real-time dispatch prone to myopic energy storage scheduling. In practice, large-scale energy storage systems (ESSs) must reserve state of charge (SoC) ahead of potential peaks, while their aggregated flexibility constraints must be evaluated within strict real-time time limits despite a large number of units. The paper proposes a proactive real-time dispatch method with fast flexibility projection for ESSs. A coarse-to-tight outer approximation of the flexibility region is constructed: a deep neural network provides an initial outer region, which is iteratively tightened by projection-based hyperplane cuts. Building on the fast flexibility projection, a reinforcement learning-based strategy is developed to optimize real-time dispatch, where a risk-aware penalty promotes proactive SoC reservation. Case studies on IEEE 118-bus systems show that the method reduces the daily peak-load shortage risk penalty by 1.17 million CNY and increases the supporting energy of ESSs by 522.04MWh, while generating real-time ESS dispatch decisions within 5.06 s.