Data-driven Wasserstein distributionally robust optimization for refinery planning under uncertainty

This paper addresses the issue of refinery production planning under uncertainty. A data-driven Wasserstein distributionally robust optimization approach is proposed to optimize refinery planning operations. The uncertainties of product prices are modeled as an ambiguity set based on the Wasserstein metric, which contains a family of possible probability distributions of uncertain parameters. Then, a tractable Wasserstein distributionally robust counterpart is derived by using dual operation. Finally, a case study from the real-world refinery is performed to demonstrate the effectiveness of the proposed approach. The results show that compared with the traditional stochastic programming method, the data-driven Wasserstein distributionally robust optimization approach is less sensitive to variations of product prices, and provides optimal solutions with better out-of-sample performance.