基于决策偏好的多目标粒子群算法及其应用

Multi-objective particle swarm optimization based on decision preferences and its application

针对传统多目标粒子群算法在解决复杂多目标优化问题上的不足,提出一种基于决策偏好的交互式多目标粒子群算法。该算法考虑决策者的正偏好和负偏好对粒子的引导作用,首先计算外部种群粒子与双极偏好点的相对贴近度,并进行排序;根据排序结果进行外部种群管理和全局最优解更新;使用δ-邻域值控制Pareto解集的分布性。在随机多目标库存控制应用中,证明了该算法对复杂应用问题求解的有效性,性能对比结果表明,该算法的收敛性、多样性和运算时间优于基于参照点的第二代非支配解排序遗传算法。

To overcome shortcomings of traditional multi-objective particle swarm methods in dealing with complicated multi-objective optimization problems,an interactive Multi-Objective Particle Swarm Optimization（MOPSO） algorithm was presented based on decision preferences.This algorithm considered the guiding roles for particles by the bipolar preferences of decision makers.The nearness degrees of out-archives particles to bipolar preferences were calculated and sequenced according to Technique for Order Preference by Similarity to an Ideal Solution（TOPSIS） method.Then,out-archives management and global-best solution update were implemented according to the sorting results,and the spreading of Pareto solution set was controlled by the-neighborhood.By applying the algorithm in stochastic multiple-target inventory control,it was proved to be effective in dealing with complicated application problems.Also,the comparison result showed that this algorithm outperformed Reference Non-dominated Sorting Genetic Algorithms II（R-NSGA-II） in convergence,diversity,and computing time.