基于自适应ε占优的多目标差分演化算法

求解多目标优化问题最重要的目的就是获得尽可能逼近真实最优解和分布性良好的非支配解集. 为此,本文提出了一种基于自适应ε占优的正交多目标差分演化算法,该算法具有如下特征:1.利用正交设计和连续空间的量化来产生具有良好分布性的初始演化种群,不仅能降低算法的时间复杂度,也能使演化充分利用种群中的个体; 2.采用在线Archive种群来保存算法求得的非支配解,并用自适应的ε占优更新Archive种群,以自适应的方式维持种群的多样性、分布性. 最后通过5个标准测试函数对算法的有效性进行了测试,并与其他的一些多目标优化算法进行了对比,实验结果显示,算法能够很好地逼近Pareto前沿,并具有良好的分布性.
The purpose to solve multi-objective optimization is to get solutions closing to the true Pareto front as much as possible and having good diversity. To meet these two demands,an algorithm is proposed in this paper,which has these characteristics:firstly,it adopts the orthogonal design method with quantization technology to generate initial population whose individuals are scattered uniformly over the target search space. So the algorithm can use them sufficiently in the subsequent iterations. What’s more,it is based on an adaptive ε concept to obtain a good distribution along the true Pareto-optimal solutions. Finally,experiments on five benchmark problems with different features have shown that this algorithm does well not only in distribution,but also in convergence when compared to other evolution algorithms