Prediction-Free Coordinated Dispatch of Microgrid: A Data-Driven Online Optimization Approach

The integration of renewable energy sources (RES) into microgrids poses challenges to reliable and economic operation due to the inherent uncertainty and volatility of RES. Contrary to the previous dispatch methods that require precise predictions of RES, this paper proposes a novel prediction-free and data-driven coordinated dispatch framework for reliable microgrid operations. In the offline stage, ex-post optimal dispatch sequences are generated based on historical dispatch from "God' s-eye view". The sequences offer a global reference and are sequentially updated based on the newly observed data. Subsequently, we propose an adaptive virtual-queue-based online convex optimization (OCO) method to generate the real-time control policy of microgrid, which aim to minimize the instant operation cost while tracking the offline reference. We provide theoretical proof that the proposed method outperforms the existing OCO methods and admits sublinear dynamic regret bound and sublinear hard cumulative constraint violation bound for OCO with time-varying constraints. Case study illustrates that the proposed method outperforms state-of-the-art methods in terms of economic optimality, computational efficiency, and security.