进化迁移优化算法综述

Survey of Evolutionary Transfer Optimization Algorithms

进化算法是模拟自然界生物进化的启发式算法,具有良好的搜索能力和灵活性且广泛用于复杂优化问题的求解,但在求解过程中默认问题先验知识为零,然而由于问题很少孤立存在,解决单一任务积累的经验可迁移至其他相关任务。进化迁移优化算法利用相关领域的知识学习和迁移,实现了更好的优化效率和性能。介绍进化迁移优化算法的基本分类,从源任务选择、知识迁移、缩小搜索空间差异、进化算法搜索、进化资源分配等5个角度出发对主流进化迁移优化算法的核心策略和优劣势进行梳理和分析。通过中国知网和WOS平台对2014年至2021年的进化迁移优化相关文献进行检索,运用知识图谱进行数据挖掘、信息处理、知识计量和图形绘制,根据进化迁移优化的发展趋势和经验分析总结了其面临的主要挑战和未来研究方向。

Evolutionary Algorithm(EA)are heuristic algorithms that simulate natural biological evolution and exhibit good searchability and flexibility.They have been successfully applied to solve complex optimization problems.However,while solving the problem,EA default the prior knowledge to zero.However,as target problems seldom exist in isolation,experience gained from a task can be transferred to other related tasks.Evolutionary Transfer Optimization(ETO)algorithms utilize knowledge learning and transferring in related fields to achieve improved optimization efficiency and performance.This study introduces a basic classification of ETO algorithms.The core strategies,advantages,and disadvantages of the mainstream ETO algorithms are sorted out and analyzed from five perspectives:source task selection,knowledge transfer,narrowing the search space difference,evolutionary algorithm search,and evolutionary resource allocation.Relevant papers on ETO published from 2014 to 2021 are retrieved through the China National Knowledge Infrastructure(CNKI)and Web of Science(WOS).The knowledge graph is used for data mining,information processing,knowledge measurements,and graph drawing.Based on the development trend of ETO and experience,the main challenges and future research directions are summarized.