基于鲁棒贡献度分类决策的大规模鲁棒进化多目标优化研究

Robust evolutionary multi-objective optimization (robust EMO) aims to find a group of Pareto-optimal solutions which are robust to small perturbations. Robust EMO plays a significant role in solving the multi-objective optimization problems in industrial production and engineering design. In the research on robust EMO, the appearance of large-scale variables will give rise to a couple of problems, including the geometric growth of the search space, difficulty of determining the robust region, etc. Therefore, the research on large-scale robust EMO is very challenging. This project pays attention to large-scale robust EMO based on contribution classification strategy. First, unconstrained large-scale robust EMO problems are discussed. According to the definition of robust contribution of decision variables, large-scale variables are divided several times by the surrogate and the analysis of separability. Then the Pareto dominance-based algorithm is developed, and the effect on Pareto-optimal solutions by disturbances under different probability distributions is investigated by the theories of truncated probability distribution and parameter analysis. Second, constrained large-scale robust EMO problems are studied. The feasibility-related variables are further classified by the surrogate and the clustering analysis to complete the large-scale robust EMO algorithm. Finally, the large-scale integrated scheduling problem in oil refinery under uncertain environment is modeled and optimized by means of the algorithm designed in this project, and we expect to provide a reference for the scheduling problem in oil refineries.

鲁棒进化多目标优化（鲁棒EMO）旨在寻找一组对扰动具有鲁棒性的Pareto最优解，对有效解决工业生产、工程设计中的多目标优化问题具有重要意义。在鲁棒EMO的研究中，大规模决策变量的出现会引起搜索空间几何增长、鲁棒域难以确定等问题，因此研究大规模鲁棒EMO是一项充满挑战的工作。本项目开展基于鲁棒贡献度分类决策的大规模鲁棒EMO研究。首先，探讨无约束条件下的大规模鲁棒EMO问题，基于决策变量的鲁棒贡献度定义，利用代理模型、可分性分析等对大规模决策变量进行多次划分，设计基于Pareto支配的大规模鲁棒EMO算法，并利用截断概率分布理论和参数分析理论探讨多种干扰分布对最优解分布的影响；其次，引入约束条件，通过代理模型、聚类分析进一步划分可行性相关决策变量，完善大规模鲁棒EMO算法；最后，利用所设计的算法，对不确定环境下的炼油过程大规模集成调度问题进行建模、优化，为炼油企业实际生产调度问题提供借鉴。

鲁棒进化多目标优化旨在寻找一组对扰动具有鲁棒性的Pareto最优解，对有效解决工业生产、工程设计中的多目标优化问题具有重要意义。在鲁棒进化多目标优化的研究中，大规模决策变量的出现会引起搜索空间几何增长、鲁棒域难以确定等问题，因此研究大规模鲁棒鲁棒进化多目标优化是一项充满挑战的工作。本项目开展基于鲁棒贡献度分类决策的大规模鲁棒鲁棒进化多目标优化研究。本项目按照研究计划执行，开展基于鲁棒贡献度分类决策的大规模鲁棒进化多目标优化研究，主要包括：基于鲁棒贡献度分类决策，设计大规模鲁棒进化多目标优化算法；并利用所设计的算法，为实际应用中的大规模鲁棒优化问题提供解决方案。针对大规模鲁棒进化多目标优化这一全新的研究方向，本项目开发了基于鲁棒贡献度分类决策的大规模鲁棒进化多目标优化算法，进行了该领域的理论初探，并针对实际生产中的调度优化问题进行了应用实践。依托本项目，在IEEE TII, TCYB等领域内知名期刊、以及GECCO、IEEE SMC等领域内知名会议发表SCI论文6篇，EI论文6篇，协助培养博士研究生2名，硕士研究生3名，并在国内外学术会议作口头报告2次，出国学术访问1次，邀请进化计算领域多名知名专家来访，与海外知名学者建立起稳定的合作关系。综上，本项目顺利完成预期目标。