Real-Time Self-Dispatch of Wind-BESS-CAES Plant: An Experience Guided Optimization Approach

The real-time self-scheduling of the Wind-BESS-CAES Plant faces the dual uncertainty of wind power and electricity prices. To address the annual operation issue, this paper proposes an experience guided online optimization framework. Firstly, the optimal scheduling trajectory is solved offline based on historical data, and a deep neural network is constructed to learn the long-term energy storage reference values, with inputs integrating historical electricity price and wind power observation data at multiple time scales (hour/day/week/month); secondly, a rolling horizon optimization model is adopted, introducing a long-term energy storage deviation penalty term in the objective function, embedding the global optimization goal across time intervals into short-term domain decisions. Case study shows that this method increases the revenue by 7.1% compared to the traditional MPC policy and by 4.0% compared to the direct application of kernel regression-based comparative algorithm, effectively coordinating the complementary characteristics of the battery and CAES, and achieving efficient online decision-making for the joint plant under uncertain environments.