双父群下动态多子群的多目标果蝇优化算法

Multi-objective Fruit Fly Optimization Algorithm for Dynamic Multiple Subgroups Under Dual-parent Group

为将果蝇优化算法(FOA)广泛应用于多目标优化问题,文章提出一种双父群下动态多子群的多目标果蝇优化算法(AMOFOA)。该算法根据解的类型划分为可行父群与不可行父群,可行父群实现Pareto解集的搜索,不可行父群实现不可行个体向可行个体的转化。针对可行父群引入搜索状态度量指标策略,根据搜索状态选择搜索操作,防止算法陷入局部最优。引入混合排序策略、领导个体选择策略,保证算法多样性,引入步长自适应更新策略动态产生子群,保证算法收敛性。同时引入存档精英策略、选择策略,实现所求Pareto解集的收敛性,引入全局外部存档删除策略,改善所求Pareto解集的多样性。针对不可行父群引入排序策略、领导个体选择策略,进一步增强算法多样性,引入步长自适应更新策略动态产生子群,保证算法收敛性。通过对标准测试函数ZDT1~ZDT3进行仿真实验,全面验证了算法AMOFOA的有效性。

In order to apply fruit fly optimization algorithm(FOA) to multi-objective optimization, this paper proposes a multi-objective fruit fly optimization algorithm(AMOFOA) for dynamic multi-subgroup under two-parent group. The algorithm is divided into feasible parent group and infeasible parent group according to the type of solution. The former realizes the search of Pareto solution set, while the latter realizes the transformation of infeasible individuals to feasible individuals. The search state metric strategy is introduced for feasible parent group, and the search operation is selected according to the search state to prevent the algorithm from falling into local optimum. The hybrid sorting strategy and the leader individual selection strategy are introduced to ensure the diversity of the algorithm, and the step-size adaptive updating strategy is introduced to dynamically generate subgroups to ensure the algorithm convergence. Meanwhile, the archiving elite strategy and selection strategy are introduced to realize the convergence of Pareto solution set, and the global external archiving deletion strategy is introduced to improve the diversity of Pareto solution set. For the infeasible parent group, the sorting strategy and the leader individual selection strategy are introduced to further enhance the algorithm diversity, and the step-size adaptive updating strategy is introduced to dynamically generate subgroups to ensure the algorithm convergence. Finally, the effectiveness of the algorithm AMOFOA is verified by simulation experiments on the standard test functions ZDT1~ZDT3.