结合分解技术的多目标引力搜索算法

Multi-objective gravitational search algorithm based on decomposition

摘要:

针对基于分解的多目标遗传算法在解决多目标问题时无法有效解决前沿面非均匀、不连续的问题,提出一种基于分解技术的多子群串行搜索的多目标引力搜索算法(MOGSA／D)．为充分利用算法优化分解出的目标函数所得到的进化信息、提高收敛速度,采取多种群串行的搜索方式;针对理想前沿面为非超平面的情况,提出一种预测理想前沿面形状的方法,并针对预测结果选择适合的权重系数生成方式;为提高解集的整体质量,提出一种基于目标权值的策略删减种群．通过标准测试函数的实验验证,所提算法与其他多目标进化算法相比在解集的收敛性以及分布性上均有较大提高,验证了算法的有效性．

Abstract:

When the ideal frontier is discontinuous or inhomogeneous, the multi-objective evolutionary algorithm can’t solve multi-objective problems effectively by decomposition. In order to improve this situation, a novel multi-objective gravitational search algorithm based on decomposition (MOGSA／D) is proposed. In MOGSA／D, the multi-population serial strategy is good for the population study evolutionary information. According to shape prediction of ideal frontier, a suitable generation method of weight coefficient is selected. A pruning strategy is adopted to prune the solution set. Experimental results show that MOGSA has a good performance to solve multi-objective problems in comparison with other multi-objective optimization algorithms.