Transferable Preference Learning Assist Multi-Objective Decision Analysis for Hydrocracking

Hydrocracking is a complex and time-consuming chemical process converting heavy oil fractions into various valuable products with low billing points, which plays a pivotal role in improving the quality of products in the oil refining process. Therefore, designing efficient surrogate models to simulate the hydrocracking process and finding appropriate solutions for multi-objective oil refining process become a research topic. To this end, we propose a novel transferable preference learning driven evolutionary algorithm for the multi-objective decision analysis in the oil refining process. Specifically, we firstly take account of the user preference to partition the objective space into a region of interest (ROI) and other subspaces, and then adopt Kriging models to approximate the sub-problems in the ROI. After that, we transfer the mutual information between the sub-problems in the ROI to enhance the robustness and generalization ability of the Kriging models during the evolutionary process. The results of the experiments on the the oil refining process have showcased the efficiency and effectiveness of our proposed approach in the multi-objective optimization and decision analysis.