Adding more renewable energy to the power system, it brings many uncertainties in generation and load that makes traditional dispatch optimization hard. This paper puts forward a dispatching optimization method based on uncertainty of power system by integrating spatiotemporal features with deep reinforcement learning. First we establish multi-time-scale dispatch model to include wind, solar production and load demand uncertainties. 2nd GC-LSTM for obtaining spatiotemporal correlations from Renewable Energy Output; 3rd Design Attention Based MA- DDPG Algorithm for Smart Dispatch Decision Making; Finally simulation results indicate that the enhanced IEEE118-node system can confirm the efficacy of the suggested approach. The result indicates that this approach can precisely capture the spatio-temporal coupling of the output from renewable energy sources, resulting in a 12.3% reduction in operating costs as compared to conventional stochastic optimization methods, along with a decrease in wind and solar curtailment rates by 35.6% as opposed to traditional approaches, and it is faster by 42.8%, with strong technical support to ensure safe and economic operation at higher levels of renewable energy.
