基于迁移学习的双层单目标优化算法

Bilevel single-objective optimization algorithm based on transfer learning

提出一种基于迁移学习的双层优化算法(BLOA-TF)来求解双层单目标优化问题(BLSOPs),该算法融合了机器学习领域的迁移学习思想.首先,通过聚类算法挑选出有代表的个体进行下层优化,将获得的下层优化信息使用归档集记录;然后,将归档集记录的下层优化信息迁移给其他相近未经下层优化的个体,以此加速整个优化过程,并有效减少计算开销;最后,将所提出的算法与通常基于嵌套的双层优化算法在12个标准测试问题上进行比较,实验结果证明了所提算法用于处理双层单目标优化问题的有效性.

A bilevel optimization algorithm based on transfer learning(BLOA-TF)was proposed to solve bilevel single-objective optimization problems(BLSOPs),which integrated the idea of transfer learning in the field of machine learning.First,the representative individuals were selected by a clustering algorithm to perform the lower-level optimization,and the obtained lowerlevel optimization information was recorded by an archive set.Then,the recorded lower-level optimization information was transferred to other individuals without performing the lower-level optimization so that the whole optimization process was accelerated with the computing cost effectively reduced.Finally,the proposed BLOA-TF was compared with a traditional nested bilevel optimization algorithm on 12 standard test problems,and experimental results validated the effectiveness of the proposed BLOA-TF for dealing with BLSOPs.