1.China Electric Power Research Institute (Nanjing), Nanjing 210003, China;2.China Electricity Council, Beijing100761, China;3.China Electric Power Research Institute, Beijing 100192, China;4.State Grid Gansu Electric Power Company, Lanzhou 730046, China

With the increasing proportion of integrated renewable energy, the significant errors in ultra-short-term forecasting remain a core challenge for dispatchers, making intra-day power balance more difficult. The hybrid look-ahead dispatch framework driven by reinforcement learning and stochastic optimization improves the adaptability of intra-day dispatch plans to uncertain scenarios through multi-scenario simulation and rolling-horizon optimization, and has been implemented in an actual provincial power grid. To meet engineering application requirements, this paper first elaborates on the decision-making paradigm and technical challenges of look-ahead dispatch. Subsequently, three practical key technologies are proposed: 1) massive power grid operational scenario generation within look-ahead horizon and balance assessment, 2) reinforcement learning-based intelligent multi-scenario simulation, 3) look-ahead decision-making considering both intra-provincial regulation and inter-provincial trading. Then, a sequential “forecast-simulation-decision” software architecture is designed and a model-data interactive-driven look-ahead dispatch technology framework is established. Finally, the test using operation data from Gansu Power Grid in China demonstrates the computational efficiency of the look-ahead dispatch software and analyzes its enhancement effects on power balancing capabilities under challenging scenarios of “ensuring supply security and promoting renewable energy accommodation”.