文献类型：会议论文

英文题名：Marine Fuel Oil Blending Formulation Control Based on Deep Reinforcement Learning

作者：Shi, Linlin[1,2,3];Chen, Ligen[1,2,3];Zhao, Yunmeng[1,2,3]

机构：[1]East China Univ Sci & Technol, State Key Lab Ind Control Technol, Shanghai 200237, Peoples R China;[2]East China Univ Sci & Technol, Minist Educ, Key Lab Smart Mfg Energy Chem Proc, Shanghai 200237, Peoples R China;[3]Huzhou Inst Ind Control Technol, Huzhou 313099, Peoples R China

会议论文集：44th Chinese Control Conference-CCC-Annual

会议日期：JUL 28-30, 2025

会议地点：Chongqing, PEOPLES R CHINA

语种：英文

外文关键词：Marine fuel oil blending; Formulation optimization; Sparse reward; Soft actor-critic; Reinforcement learning

摘要：The online optimization of marine fuel oil blending could bring significant economic benefits to refineries, especially in profiting from heavy oil resources. However, marine fuel oil blending optimization faces challenges, including nonlinear mixing features of properties like kinematic viscosity, and fluctuations in component oil properties. Herein, we propose an online optimization method for marine fuel oil blending based on a deep reinforcement learning (DRL) framework. The method introduces techniques such as mapping property constraints and scaling factors to resolve the sparse reward problem during the training of the DRL. The Soft Actor-Critic (SAC) algorithm, which offers high training stability and data efficiency, adjusts the blending recipes for marine fuel oil. Compared to a traditional blending recipe optimization method, the proposed approach demonstrates higher economic benefits and exhibits greater robustness under fluctuations in component oil properties.