带偏向性轮盘赌的多算子协同粒子群优化算法  

A multi-operator collaborative particle swarm optimization algorithm with biased roulette

作　　者：于海波 朱秦娜 康丽 乔钢柱 曾建潮 YU Hai-bo;ZHU Qin-na;KANG Li;QIAO Gang-zhu;ZENG Jian-chao(College of Computer Science and Technology,North University of China,Taiyuan 030051,China;Institute of Big Data and Visual Computing,North University of China,Taiyuan 030051,China;College of Environment and Safety Engineering,North University of China,Taiyuan 030051,China)

机构地区：[1]中北大学计算机科学与技术学院,太原030051 [2]中北大学大数据与视觉计算研究所,太原030051 [3]中北大学环境与安全工程学院,太原030051

出　　处：《控制与决策》2024年第4期1167-1176,共10页Control and Decision

基　　金：国家自然科学基金青年基金项目(62106237);国家自然科学基金联合基金项目(U21A20542);山西省自然科学基金项目(201901D211237)。

摘　　要：针对粒子群优化算法在处理高维、大规模、多变量耦合、多模态、多极值属性优化问题时易早熟收敛等性能和技术瓶颈,基于粒子群优化算法行为学习算子和3种不同学习偏好的差分变异算子,建立带偏向性轮盘赌的多算子选择与融合机制,提出一种带偏向性轮盘赌的多算子协同粒子群优化算法MOCPSO.MOCPSO针对迭代粒子群榜样粒子集,首先通过对迭代种群及其榜样粒子集优劣分组,同时采用轮盘赌分别为每组榜样粒子集选配不同学习偏好的变异算子,并为每组榜样粒子适配差分基向量和最优基向量,预学习并优化迭代种群及其榜样粒子,以权衡算法的全局探索和局部开发;然后通过合并所有子种群,并结合粒子群优化算法行为学习算子,指导迭代种群状态更新,以提高算法的全局收敛性;最后结合精英学习策略,对群体历史最优进行高斯扰动,以提高算法的局部逃生能力,保障算法收敛的多样性.实验结果表明,MOCPSO算法与5种先进的同类型群智能算法在求解CEC2014基准测试问题上具备竞争力,且有更强的优化特性.To address the performance and technical bottlenecks of a particle swarm optimization algorithm in tackling optimization problems of high-dimensional,large-scale,multivariate coupling,multi-modal,multi-extreme attribute vulnerable to premature convergence,a multi-operator selection and fusion mechanism with biased roulette is established based on the behavioral learning operator of particle swarm optimization and three differential mutation operators with different learning preferences and a multi-operator collaborative particle swarm optimization algorithm with biased roulette is proposed(MOCPSO).For balancing the exploration-exploitation trade-off,MOCPSO first groups the iterative swarm into several subswarms with different learning tasks according to the fitness,where each subswarm configures a differential mutation operator that the differential mutation vectors selected among the exemplars of all subswarms through roulette selection,to pre-learn and optimize the iterative swarm and their exemplar particles.Then all subswarms are merged undergoing behavioral learning operation of the particle swarm optimization to improve the global convergence.Finally,for guaranteeing the diversity of algorithm convergence,the MOCPSO incorporates an elitist learning strategy to guide iterative swarm escaping the possible local traps by performing Gaussian perturbation on the current global best.Experimental results show that the proposed MOCPSO algorithm possesses stronger and more competitive optimization properties than five state-of-the-art swarm intelligence algorithms in solving the CEC2014 benchmark test suit.

关 键 词：粒子群优化 差分演化 多算子协同 榜样竞争 偏向性变异策略 精英学习 

分 类 号：TP18[自动化与计算机技术—控制理论与控制工程]