基于种群多样性控制的多级信息迁移多任务优化粒子群算法  

Multi-level information transfer multi-task PSO based on population diversity control

作　　者：程美英 钱乾 倪志伟[3] CHENG Mei-ying;QIAN Qian;NI Zhi-wei(School of Economics&Management,Huzhou University,Huzhou 313000,China;School of Teacher Education,Huzhou University,Huzhou 313000,China;School of Management,Hefei University of Technology,Hefei 230009,China)

机构地区：[1]湖州师范学院经济管理学院,浙江湖州313000 [2]湖州师范学院教师教育学院,浙江湖州313000 [3]合肥工业大学管理学院,合肥230009

出　　处：《控制与决策》2024年第3期728-738,共11页Control and Decision

基　　金：国家自然科学基金青年基金项目(62102148)。

摘　　要：基于群体智能“隐并行性”实现多任务优化已取得一系列研究成果,但任务间频繁的垂直信息传递导致种群异质性过度增加,进而产生信息负迁移消极影响,这也是目前多任务优化领域尚未完全解决的难题之一.针对此问题,首先将粒子群算法(PSO)与多种群演化信息共享机制相结合,然后引入标杆管理思想实现多层级信息迁移及智能涌现,最后通过计算种群多样性指数有效控制信息迁移频率,提出多级信息迁移多任务优化PSO算法(multi-level information transfer multi-task PSO, MITMPSO).仿真实验表明,通过设置合理的信息迁移阈值,MITMPSO能在多项式时间内显著提高多任务高维函数优化、多任务多约束函数优化以及多任务二元离散优化问题的求解质量,加快各优化问题的收敛速度.A series of research achievements have been made in multi-task optimization(MTO)based on the implicit parallelism of swarm intelligence.However,the frequent vertical information transfer between tasks leads to excessive increase of population heterogeneity,resulting in the negative impact of information migration,which is also one of the problems that has not been completely solved in the field of MTO.Firstly,PSO and multi-population evolution information sharing mechanism are combined,then the idea of benchmarking management is introduced to realize multilevel information migration and intelligent emergence,finally the frequency of information transfer is effectively controlled by calculating the population diversity index,and the multi-level information transfer multi-task PSO(MITMPSO)is proposed.Experimental results show that the MITMPSO can significantly improve the solution quality and accelerate the convergence speed of multiple high-dimensional functions,multiple multi-constraints functions and multiple binary discrete optimization problems concurrently in polynomial time by setting a reasonable information migration threshold.

关 键 词：多任务优化 粒子群算法 多级信息迁移 种群多样性控制 

分 类 号：TP18[自动化与计算机技术—控制理论与控制工程]