Abstract: Improving the operational stability of the power grid while further enhancing the economic efficiency of energy dispatch is an important direction in the advancement of energy systems. Compared to conventional integrated energy systems, large-scale clean energy bases with gigawatt-scale installed capacity have significantly higher wind and solar power installed capacities, leading to greater output volatility and challenges in maintaining operational economic efficiency. This requires more flexible and real-time adjustable dispatch plans to enhance operational economy while ensuring stable output. Based on these factors, this study focuses on the Qinghai new energy base and proposes a multi-time-scale, two-stage optimization dispatch method, combining its installed capacity structure and resource characteristics. In the first stage of day-ahead scheduling, forecasting data for wind, solar, and load power for the following day are utilized to maximize operational economy by optimizing the output scheduling of controllable generation units. In the second stage of real-time optimization scheduling, adjustments are made the output power of hydropower and energy storage controllable units based on deviations between actual wind and solar power outputs and the dispatch plan. This adjustment employs a predictive control model (model predictive control, MPC) based on LSTM predictions to mitigate the economic impact of forecasting errors. Finally, actual data from typical days in Qinghai are simulated, and this two-stage scheduling model is compared with a scheduling strategy. In this manner, the two-stage scheduling model has been validated to have the capability of effectively improving the operation economy while maintaining system supply and demand balance.