论文部分内容阅读
提出了一种基于分解的多目标差分进化算法(MODE/D).该算法使用差分进化算法替代遗传算法作为搜索引擎,采用混料均匀设计方法确定种群的权重向量,使用循环式拥挤距离排序策略维护外部档案的多样性.另外,为了改善MODE/D的收敛性能,将局部搜索的思想融入到MODE/D算法中.该算法采用聚合函数作为局部搜索的优化函数,并使用SQP(sequential quadratic programming)方法进行求解.实验研究表明,相对于MOEA/D与NSGA-II,MODE/D的收敛性与多样性能均得到了改善,验证了MODE/D的有效性.最后将MODE/D算法应用到热轧带钢窜辊策略的多目标优化中,工业生产证明,优化窜辊策略显著延长了同宽轧制公里数,改善了带钢断面轮廓与板形质量.
A new multi-objective differential evolution algorithm (MODE / D) based on decomposition is proposed in this paper, which uses differential evolution algorithm instead of genetic algorithm as search engine and mixture homogeneous design method to determine the weight vector of population. Using cyclic congestion distance sorting strategy In order to improve the convergence performance of MODE / D, the idea of local search is incorporated into the MODE / D algorithm, which uses the aggregation function as the optimization function of local search and uses SQP (sequential quadratic programming ) Method.The experimental results show that the convergence and diversity of MODE / D are improved compared with that of MOEA / D and NSGA-II, and the validity of MODE / D is verified. Finally, the MODE / D algorithm is applied to In the multi-objective optimization of the hot strip rolling strategy, industrial production proves that the optimized pass-rolling strategy significantly prolongs the number of kilometer-width rolling and improves strip profile and plate shape quality.