Resource Aggregation and Optimal Dispatch of Virtual Power Plants: A Spatio-Temporal Data-Driven Approach

Under the background of large-scale renewable energy integration, virtual power plants (VPPs) realize unified coordination and information interaction between generation resources, user loads and power grids, which serves as a critical technical means for efficient operation in the new energy scenario. To improve energy utilization efficiency of VPPs, this paper proposes a data-driven optimal dispatch framework. First, cluster analysis is performed to clean massive operation data and remove redundant information. Then, an improved long short-term memory (LSTM) method is adopted to extract time-series features of wind–solar output and user load. On this basis, a resource aggregation and optimal control model is constructed using real-time demand response and consensus algorithm. Case studies demonstrate that all clusters reach consistent convergence after 30 iterations with a cost increase rate of 0.28 Yuan/kW. The optimal regulation powers of clusters 4, 6, 9, 10 and 13 are 7.2, 3, 1.67, 0.43 and 0.58 kW, respectively. When cluster 14 participates in regulation temporarily, convergence steps rise but the minimum-cost power optimization remains valid. Simulation of whole-day dispatch shows that demand response effectively promotes renewable energy consumption and reduces the comprehensive operating cost by 4.86%. The proposed strategy can strengthen the renewable energy accommodation capability and reduce operation costs of VPPs.