Self-Improving Online Storage Control for Stable Wind Power Commitment

The integration of distributed energy resources, particularly wind energy, presents both opportunities and challenges for the modern electrical grid. On the supply side, wind farms frequently encounter penalties due to wind power’s intermittency and variability. The incorporation of energy storage systems can mitigate these penalties through real-time power adjustments. However, the uncertainties in future renewable generation significantly impede optimal storage control, and existing algorithms either lack theoretical guarantees, or fail to effectively leverage data to attain better performance. This paper effectively addresses this dichotomy by bridging the gap between data utilization and theoretical guarantees based on the Markov decision process. Specifically, we first introduce a one-shot online storage control algorithm that utilizes historical data to make near-optimal decisions with theoretical performance guarantees. To further enable continuous learning from new data, we develop an online learning-based self-improving storage control algorithm, underscoring its asymptotic optimality. The numerical study using field data demonstrates the efficacy of the proposed approach.