Distributionally Robust Unit Commitment Models: Theory and Numerical Results

The Unit Commitment (UC) problem seeks optimal generation schedules under technical and economic constraints. Increasing renewable energy penetration amplifies uncertainty in net load forecasting. Classical approaches, such as stochastic and robust UC models, partially address this challenge but face wellknown drawbacks: stochastic formulations may overfit sampled scenarios, while robust methods tend to yield overly conservative solutions. To overcome these limitations, Distributionally Robust Optimization (DRO) has emerged as a promising alternative by explicitly accounting for ambiguity in probability distributions. While prior DRO studies in UC rely on moment constraints, ϕ- divergences or the Wasserstein distance with the L1 norm, the potential of other Wasserstein distances, especially with the L2- norm, remains unexplored. This paper presents a unified Benders decomposition framework for two-stage UC models with righthand side uncertainty that efficiently solves various approaches, including DRO using Wasserstein distance. Numerical experiments on IEEE test systems evaluate out-of-sample performance and computational times across different uncertainty levels.