The aggregated flexibility of distributed controllable resources—including generation, load, and storage—plays a crucial role in maintaining the dynamic equilibrium of distribution grid systems. To fully leverage the regulation capabilities of these resources during real-time grid operation, there is an urgent need for real-time prediction of their flexibility, yet existing research has rarely addressed this region. This paper proposes a real-time prediction method for the aggregated flexibility of multi-source distributed controllable resources. First, a single-unit power–energy boundary model is constructed for typical distributed generation, load, and storage resources. This model unifies the characteristics of time-coupled storage resources and time-decoupled generation-load resources by superimposing power and energy boundaries and introducing a contraction coefficient. This approach simultaneously addresses errors arising from aggregation model inaccuracies and inherent uncertainties. Subsequently, an aggregation feasible region prediction model based on the Transformer architecture is designed. The prediction results are quantified using absolute and relative errors during the decomposition process. Finally, validation using operational data from a microgrid in Zhejiang demonstrates that the proposed method can predict the boundaries of distributed resource aggregated flexibility in a real-time rolling manner. Its accuracy achieves the highest prediction among the compared methods, while also controlling the degree of conservatism to meet the requirements for grid control command execution errors.
Xu et al. (Sun,) studied this question.