摘要
针对输入空间包含多种类型的数据时,以单一的神经网络为模型,其收敛很困难的问题,提出一种基于模块小波神经网络的建模方法.利用分而治之思想,模块神经网络通过一个门控网络进行分类和协调,可以将一个复杂任务分解成几个简单的子任务,每个子任务由一个局部专家网络学习.与传统的模块网络不同,这里的专家网络是小波网络而不是BP网络.将所提出的网络模型用于热连轧产品质量建模,并与单一的神经网络建模结果进行比较.建模结果表明,模块小波神经网络模型优于单一神经网络模型.
When the input space consists of several different classes of input data, it becomes very difficult to converge the network during the training phase. A modular wavelet neural-network is presented to overcome this difficulty. Based on the divide-and-conquer concept, a modular network is capable of dividing a complex task into subtasks, and modeling each subtasks with an expert network. To model such activities, a gating network is used for the classification and allocation of the input data to the corresponding expert network. Different from traditional modular networks, here each expert network is a wavelet network. The performance of such networks in modeling product quality is examined and compared with that of singular networks. Modeling results demonstrate that the proposed method takes a significant improvement over the general singular network model.
出处
《控制与决策》
EI
CSCD
北大核心
2004年第3期295-298,共4页
Control and Decision
基金
国家863计划资助项目(863-51-011)
国家自然科学基金资助项目(60274055)
西安交通大学自然科学基金资助项目(0900-5-73024).
关键词
模块小波网络
高维输入
质量模型
热连轧机
Backpropagation
Expert systems
Hot rolling mills
Quality assurance
Wavelet transforms
