按区域惩罚划分的并行多目标遗传算法

Parallel multi-objective genetic algorithm by adding area penalty

解决多学科设计优化问题的多目标遗传算法通常面临着大计算量的挑战,提出了一种新型的并行化算法来提高其效率.全局个体均匀的分布在各个进程,首先从所有的进程中获取全局范围的Pareto最优解极值,并发送给每个进程,再由这些极值来构造各个进程自己的惩罚函数.通过惩罚函数给个体添加约束来划分各个进程的收敛区域,同时采取优化措施保证每个进程加速收敛并且收敛区域没有重叠和遗漏,这样每个进程只需收敛到特定的一段Pareto最优解,降低了计算量;同时由于进程间交换的数据量小,保证了效率的提高.通过与串行算法(NSGA2)和其他的并行化算法比较,显示了该算法的有效性和先进性.

One challenge for multi-objective genetic algorithm （MOGA） is the computational cost when MO- GAs were used in the multidisciplinary optimization （MDO） problems. To improve the efficiency of MOGA, a new parallel algorithm was suggested. All the individuals were distributed among processors equally, and each processor got the extremum of Pareto solutions from all processors and constructed its own penalty function. Then each processor could divide its own Pareto solutions convergence area by the penalty function. To avoid the appearance of overlapping and omitting area and reduce the convergence time, some optimization techniques were suggested. So each processor could converge to its own special Pareto solutions segment. Because the individuals computed was divided into every processor equally, in each processor the computational cost was reduced. This with the small data changed in each processor guaranteed the efficiency. Through comparing with serial MOGA （NSGA2） and the other parallel MOGA （guided domination approach）, the algorithm is proved being more effective and advanced