一种增强型多目标烟花爆炸优化算法

摘要

现实中多目标优化问题的多样化和复杂化要求发展新的多目标优化算法.在混合多目标进化算法设计思想和新型进化模型的启发下,提出一种增强型多目标烟花爆炸算法eMOFEOA,该算法利用均匀化与随机化相结合的方法生成均匀分布的初始种群,为算法后续搜索提供较好的起始点;对烟花爆炸半径采用精细化控制策略,即不同世代的种群具有不同的爆炸半径,而且同一种群内部因个体支配强度的差异而具有不同的爆炸半径,以节省计算资源;利用简化的k-最近邻方法维持外部档案的多样性.本文算法与另5种对等比较算法一同在12个基准多目标测试函数上进行性能比较,实验结果表明eMOFEOA算法在收敛性、多样性和稳定性上具有总体上显著的性能优势.%In reality,the diversification and complexity of the multi-objective optimization problems (MOPs) require the development of some novel multi-objective optimization algorithms.Inspired by the hybrid multi-objective evolutionary algorithms (MOEAs) and new evolutionary instances,an enhanced multi-objective fireworks explosion optimization algorithm (eMOFEOA for short) is proposed to solve the hard MOPs efficiently in the paper.Firstly,the proposed approach uses the approach of combining uniformization and randomization to generate an initial population that are scattered uniformly over the feasible search space,so that the algorithm can acquire a good beginning for the subsequent iterations.Secondly,a fine control strategy of explosion radius is adopted in the eMOFEOA,that is to say,different generation of population has different radius,and the different firework in the same generation have different radius based on its strength of Pareto dominace,so as to save the computation resource to the maximum extent.Thirdly,a simplified k-nearest neighbor approach is employed to maintain the diversity of external archive in the eMOFEOA.The proposed eMOFEOA is compared with the other five peer comparison algorithms in the performance of convergence and diversity based on 12 benchmark multi-objective test functions,and the experimental results show that our eMOFEOA has the overall performance advantages in convergence,diversity and stability.