Data-Driven Park-Level Virtual Power Plant Self-Scheduling Based on the Quarterly Budget and the Corrected Conditional Expectation

The distribution energy resource (DER) is widely recognized as an essential element in future industrial systems, which assist in reducing fossil energy consumption. The aggregation of DERs yields the notion of virtual power plant (VPP), which is beneficial for the power grid dispatching. However, due to the small aggregation scale, limited flexible resources and the uncertainty of DERs in the park-level VPP (PVPP), the new uncertainty strategies are in urgent need of research. In this paper, a two-stage improved distributionally robust (DR) approach is proposed for the PVPP self-scheduling. Firstly, the PVPP self-scheduling plan is formulated to satisfy the local consumption demand and supply the rest capacities for the electricity market as much as possible for the sake of profits. When trading with electricity markets, the PVPP not only acts as a consumer or a producer to purchase or sell the energy, but also purchases the external reserve energy from reserve markets. Then, we construct the quarterly budget-based ambiguity set to dynamically determine a less conservative uncertainty budget. Further, the ridge regression algorithm (RRA) is adopted for conditional expectation estimator to correct the ambiguity set, which is used to achieve a more accurate performance during the re-dispatch. Case studies on real-world data evidence the effectiveness of the proposed PVPP DR approach in terms of robustness against system uncertainties and reducing the energy costs in industrial environments.