# [1]李二超,李进.两阶段三存档集约束优化算法（TSDA）[J].郑州大学学报(工学版),2018,39(06):23-29. 　Constraint Optimization Algorithm with Two-Stage and Three-Archive[J].Journal of Zhengzhou University (Engineering Science),2018,39(06):23-29. 点击复制 两阶段三存档集约束优化算法（TSDA）() 分享到： var jiathis_config = { data_track_clickback: true };

39

2018年06期

23-29

2018-10-24

## 文章信息/Info

Title:
Constraint Optimization Algorithm with Two-Stage and Three-Archive

A

针对约束优化算法采用相同的进化策略处理位于Pareto边缘的解与函数值较差的解，使得寻优结果较差，提出一种两阶段三存档集约束优化算法。该算法分为两个阶段。第一阶段：根据£（t）值将种群分为3个存档集，即非支配解存档集、支配解存档集以及非支配可行解存档集。非支配解存档集进行混合策略的双重寻优，既避免了算法陷入局部最优，又使得靠近前沿的解加速收敛；支配解存档集则注重于全局搜索，从而有利于算法搜索到更优可行解。非支配解存档集和支配解存档集使用不同的优化策略进行进化，提高了算法的寻优能力。第二阶段：在第一阶段达到设定的代数时，将各代保留到非支配可行解存档集中的个体进行快速非支配排序，选出的N个优秀个体则为最优解。最后，将提出的算法与其约束多目标进化算法在3种经典约束测试函数上进行对比。仿真结果表明，所提出算法在不同类约束条件下的寻优能力均具有优势。
Abstract:
Constrained optimization algorithm adopted the similav evolutionary strategy to deal with solutions located on the Pareto edge and solutions with poor function values, which could lead to poor search results. Aiming to solve this problem, a constrained optimization algorithm with teo-stage and three-archive was proposed. The algorithm was divided into two stages. In the first stage, the population was divided into three archives according to the £(t) value. These archives were non-dominated solution archives, dominant solution archives, and non-dominated feasible solution archives, respectively. The dual optimization of hybird strtegy is applied to the non-dominated  solution archives. It could not only aviod being trapped in local optimum, but also accelerate the convergence of solutions near the frontier. The dominant solution archives focused on the global search, which was benefical for the algorithm to search better feasible solution. The non-dominated solution archives and the dominant solution archives were evolved using different optimization strategies to improve the optimization capability of algorithm. In the second stage, non-dominated sorting was performed on individuals when the first stage reached the certain generation. These individuals were concentrated from each generation to the non-dominated feasible solution archives. The selected N individuals were the optimal solution. Finally, the proposed algorithm was compared with other constrained multi-objective evolutionary algorithms on the three classical constraint test functions. The simulation results showed that the proposed algorithm had advantages in different kind of constraints.