A hybrid data-driven robust optimization approach for unit commitment considering volatile wind power

This paper presents a data-driven hybrid method combining kernel density estimation (KDE) and Wasserstein metric to solve unit commitment problems considering volatile wind power. The proposed approach adopts KDE to deduce the underlying probability distribution function (PDF) of wind power forecasting errors, and utilizes the confidence interval of the estimated PDF to construct the support of uncertainties. Wasserstein metric is applied to measure the distances among distributions belonging to the established support. We take the PDF estimated by KDE as the center, and take a certain Wasserstein distance, so as to form a Wasserstein ball of probability distributions, namely ambiguity set. The unit commitment plan is obtained by minimizing the expected cost with regard to the worst-case distribution in the ambiguity set. As a fusion of Wasserstein technique and KDE, the proposed method inherits good abilities of conventional Wasserstein-based method, such as statistics, tractability, extensibility and data mining capacity. More importantly, since the PDF estimated by KDE can converge to the true distribution with the increase of sample data, the proposed method can effectively further narrow the conservatism of the model without losing robustness. Computational results on IEEE 57-bus, 118-bus and 145-bus systems from MATPOWER 6.0 numerically demonstrate the favorable features of the proposed method.