Data-Driven Bayesian Nonparametric Wasserstein Distributionally Robust Optimization

Abstract: In this work, we develop a novel data-driven Bayesian nonparametric Wasserstein distributionally robust optimization (BNWDRO) framework for decision-making under uncertainty. The proposed framework unifies a Bayesian nonparametric method and the Wasserstein metric to decipher the global-local features of uncertainty data and encode these features into a novel data-driven ambiguity set. By establishing the theoretical connection between this data-driven ambiguity set and the conventional Wasserstein ambiguity set, we prove that the proposed framework enjoys the finite sample guarantee and asymptotic consistency. To efficiently solve the resulting distributionally robust optimization problem under the BNWDRO framework, we derive for this optimization problem an equivalent reformulation, which is kept tractable for many practical scenarios. Numerical experiments on a unit commitment problem verify the effectiveness of the proposed BNWDRO framework compared with existing methods.