Data-driven robust optimization with multiple kernel learning for refinery planning under price uncertainty

Abstract

Refinery planning is crucial for increased profitability for refineries. However, the markets associated with refinery operations are volatile, resulting in fluctuations in the product price, which can heavily affect the total profit of refineries. This paper is intended to develop a data-driven robust optimization (DDRO) framework for refinery planning under price uncertainty. Firstly, historical data of the product prices is collected and a multiple kernel learning (MKL) algorithm is proposed to construct the uncertainty set to capture the price uncertainty. Then, based on the derived uncertainty set, a DDRO model of refinery planning is developed and a tractable robust counterpart is reformulated by using the dual transformation, which is directly solved by using the solver. Finally, an industrial case of refinery planning is researched to illustrate the applicability of the proposed approach, which demonstrates that the proposed approach has a better balance between the total profit and robustness for refinery planning than the deterministic method.