Abstract: Firstly, based on the four-stage closed-loop framework of “data perception–scenario generation–planning optimization–operation verification”, multi-source data including smart meters, PMUs, and meteorological systems are integrated. The LSTM and Transformer models are adopted to realize operation state perception and short-term load forecasting. Then, typical scenarios of renewable energy generation are produced using Latin hypercube sampling and K-means clustering. Aiming at low network loss and high consumption, collaborative optimization of power grid topology and renewable energy access capacity is realized based on NSGA-II. Finally, combined with probabilistic power flow, continuous power flow, and MPC-based coordination verification of energy storage and flexible loads, the closed-loop feedback between planning and operation is accomplished. A case study of a regional power grid shows that the annual curtailment rate is reduced from 11.3% to 1.87%, and the voltage violation probability is reduced from 12.7% to 2.14%, effectively improving renewable energy accommodation and system security.