摘要
针对传统多目标优化算法在其领域存在的多个子目标不能同时取优的问题,提出了一种基于改进的非支配排序遗传算法(non-dominated sorting genetic algorithm-Ⅱ,NSGA-Ⅱ)多目标优化方法。以多目标优化遗传算法为基础,多输入多输出的反向传播(back-propagation,BP)神经网络为适应度函数评价体系,保证算法快速收敛并搜索到全局最优解集。该算法在建模前对实验数据进行主成分分析,降低了运算时间和算法难度,通过在遗传进化过程中引进正态分布交叉算子(normal distribution crossover,NDX)和改进的自适应调整变异算子,实现了多个目标同时取优,保证Pareto最优解集快速、准确地获取。仿真实验使用UCI数据集,通过与其他常用的多目标优化算法对比,验证了改进的NSGA-Ⅱ算法精确度更高、收敛速度更快、稳定性更强。
The traditional algorithm cannot optimize the subgoals at the same time in multi-objective optimization area. To solve this problem,on the basis of improved NSGA-Ⅱ,this paper presented a kind of multi-objective optimization method. On the foundation of multi-objective optimization genetic algorithm,it took the BP artificial neural network model of multi-inputsmulti-outputs as fitness evaluation function for NSGA-Ⅱ. So that the algorithm could converge on a fast speed and found out the global optimal solutions. In use of reducing dimensions,this paper analyzed the principal component of the experimental data before system modeling,thus it could reduce the running time and algorithm difficulty. To realize the aim of multi-objective optimal at the same time,the paper proposed the NDX operator and the improved adaptive adjustment mutation operator in genetic evolution process. So this experiment can obtain Pareto optimal solution set quickly and accurately. Compared with the conclusion of other multi-objective optimization algorithms,in the experimental simulation with UCI standard data set,the result of improved NSGA-Ⅱ shows that it has higher precision,the faster convergence and the stronger stability.
作者
路艳雪
赵超凡
吴晓锋
韩晓霞
Lu Yanxue;Zhao Chaofan;Wu Xiaofeng;Han Xiaoxia(College of Information Engineering,Taiyuan University of Teehnology,Jinzhong Shanxi 030600,Chin)
出处
《计算机应用研究》
CSCD
北大核心
2018年第6期1733-1737,共5页
Application Research of Computers
基金
国家青年科学基金资助项目(21606159)
关键词
降维
搜索空间
遗传算子
神经网络
多目标优化
非支配解
dimensionality reduction
searching space
genetic operator
neural network
muhi-objeetive optimization
nondominated solution
