摘要
给出了多变量复杂系统的遗传 神经网络建模问题的学习训练法.该方法利用竞争学习对输入空间进行聚类,并对初步网络权值利用遗传算法进行修剪,从而优化网络拓扑结构.然后对优化网络进行训练学习,得到最终模型.最后以文献资料的例证为例,应用该神经网络估算模型进行验证.结果表明,该模型估算准确,由于考虑了变量的最优选择,因此非常适应于工程类复杂系统的动态管理.
The paper proposed a new modeling method for multivariable complexity system. The method is applied in a competition learning by the means of clustering input variable space, and prunes the network weight based on the genetic algorithm in the purpose of getting an optimum topologic structure. And then the prototype was being trained by the neural network. While the learning is over, we get the final optimum model. Experiment shows that it is an effective method in the engineering management.
出处
《天津理工学院学报》
2004年第1期35-38,共4页
Journal of Tianjin Institute of Technology
基金
天津市高等学校科技发展基金资助项目(020725)