摘要
为解决后验式场景下的多目标生产调度问题,提出一种基于自组织映射神经网络的策略来生成近似Pareto边界。该方法首先使用拉格朗日松弛法获得若干Pareto解,从而将搜索范围划分为若干区域。对于每一个区域,构造两个并发运行的自组织映射神经网络搜索区域中的Pareto解,在不增加求解时间的情况下提高了求解精度。另外,根据多目标调度问题的特点,改变了神经网络训练过程中邻域的定义,从而加快了求解速度。仿真实验验证了该算法的可行性与有效性。
To solve the posterior multi-criteria scheduling problems, a new method based on Self-Organization feature Map neural network （SOM） was presented to generate approximate Pareto boundaries. Firstly, Lagrange relaxation algorithm was applied to obtain several Pareto optima which were exploited to divide the search space into several domains. For each domain, two simultaneously running SOM were constructed to explore the remaining Pareto optima. According to the char- acteristics of multi-criteria scheduling, a new definition of neighbor for neural network training was proposed. Numerical experiments demonstrated the feasibility and the effectiveness of this algorithm.
出处
《计算机集成制造系统》
EI
CSCD
北大核心
2008年第4期757-760,820,共5页
Computer Integrated Manufacturing Systems
基金
国家自然科学基金资助项目(60474002
60504026)
国家863计划资助项目(2006AA04Z173)~~
关键词
多目标调度
Pareto边界
自组织映射神经网络
拉格朗日松弛法
multi-criteria scheduling
Pareto boundary
self-organizing feature map neural network
Lagrange relaxation algorithm
