Distributionally Robust Optimization for Medium-Voltage Distribution Network Operation Considering Source-Load Error Observability Differences

Abstract

With the increasing penetration of distributed photovoltaics, electric vehicles, and other distributed energy resources, the net load of medium-voltage distribution networks exhibits stronger volatility, while its forecast-error distribution is difficult to characterize accurately under limited historical samples. Conventional distance-based distributionally robust optimization (DRO) generally adopts a uniform ambiguity-set radius, ignoring the differences in the representativeness of historical error samples across nodes and time periods. To address this problem, this paper proposes a DRO operation method considering source-load error observability differences. The proposed method first constructs an empirical distribution using historical forecast-error samples of nodal net loads. Then, a normalized high-quantile forecast-error index is introduced to quantify error-tail deviations, and a source-load error observability correction coefficient is formulated accordingly. On this basis, the Wasserstein ambiguity-set radius is adaptively corrected to provide larger risk margins for nodes and periods with significant error-tail deviations. The corrected ambiguity set is embedded into the distribution-network DRO operation model. Case studies demonstrate that the proposed method improves the balance between operational economy and security compared with deterministic optimization and conventional DRO with a uniform ambiguity-set radius.