Diesel blending under property uncertainty: A data-driven robust optimization approach

•A DDRO model for diesel blending optimization is developed.•The uncertainty set is constructed using the PCA-RKDE method.•A case study from an actual diesel blending system is carried out. With the increasing demand for diesel fuel and the strict diesel standards, diesel blending has become an essential technology in response to clean energy. However, in the traditional blending process, the operator does not consider the uncertainty of component oil properties during the recipe optimization, resulting in a sub-optimal or even infeasible solution. This paper proposes a data-driven robust optimization framework to address this issue. First, a hybrid machine learning method combining principal component analysis and robust kernel density estimation is used to construct uncertainty sets to capture uncertain properties. Then, a data-driven diesel blending model is formulated using the derived uncertainty set through the dual operation. Last, an actual case study is implemented to show the effectiveness of the proposed method in handling uncertainties and obtaining a good balance between robustness and optimality of the recipe optimization for diesel blending. Moreover, the parameters of the uncertainty sets are analyzed in detail for providing reasonable parameters to guide the actual diesel blending.