Abstract: Aiming at the problems of lacking of convergence and loss of diversity in Multi-Objective Particle Swarm Optimization(MOPSO)algorithm,an adaptive multi-objective particle swarm optimization algorithm(pmdMOPSO)based on evolutionary state equilibrium convergence performance and diversity energy was proposed.The algorithm firstly adopted the Manhattan distance real-time detection the evolutionary state of algorithm,divided the evolution state into two stages of exploration and convergence,and adopted different speed update modes respectively according to the two stages of the evolution state,to improve the performance of the algorithm.Secondly,a velocity dynamic equation with Levy flight exploration cognitive behavior was designed to enhance the overall exploration ability.Finally,the adaptive update model of evolutionary parameters was designed by using the differential vector composed of the population Manhattan distance to balance the ability of the algorithm for global exploration and local mining.The experimental results on the MOP1-MOP7 test function showed that the pmdMOPSO algorithm had better convergence performance and diversity than the comparison algorithm.

Key words: multi-objective particle swarm optimization algorithm, population Manhattan distance, Levy flight exploration knowledge, parameter adaptation

摘要： 针对多目标粒子群优化算法存在收敛性不足和多样性丢失问题,提出一种根据进化状态平衡收敛性能与多样性能的自适应多目标粒子群优化算法(pmdMOPSO)。该算法首先采用种群曼哈顿距离实时检测算法的进化状态,将进化状态分为探索和收敛两个阶段,并根据进化状态的两个阶段分别采用不同的速度更新模式,实现算法性能的提升;其次设计了具有Levy飞行探索认知行为的速度动力学方程,旨在增强全局探索能力;最后使用种群曼哈顿距离构成的差分向量设计进化参数自适应更新模式,平衡算法全局探索与局部开采的能力。通过对MOP1~MOP7测试函数上的实验结果分析,表明pmdMOPSO算法较对比算法具有更好的收敛性能和多样性能。

关键词: 多目标粒子群优化算法, 种群曼哈顿距离, Levy飞行探索认知, 参数自适应