基于灰色关联分析的高阶神经网络剪枝算法  

Pruning algorithm training high-order neural network based on grey incidence analysis

作　　者：熊焱[1,2] 吴微[1] 张超[1] 亢喜岱[1] 

机构地区：[1]大连理工大学数学科学学院,辽宁大连116024 [2]辽宁科技大学理学院,辽宁鞍山114051

出　　处：《大连理工大学学报》2010年第3期463-468,共6页Journal of Dalian University of Technology

基　　金：国家自然科学基金资助项目(10471017)

摘　　要：在综合分析网络纵向、横向灰色关联分析特点的基础上提出了一种新的基于灰色关联分析的剪枝算法,并将其用于训练高阶神经网络.该算法运用灰色关联分析对比网络各节点输出值序列之间联系的紧密程度,用网络纵向灰色关联分析确定剪枝连接,再用网络横向灰色关联分析确定相应的并枝连接,实现网络结构的动态修剪.训练后的高阶神经网络具有合理的网络拓扑结构和较好的泛化能力.实验验证了该算法的合理性、有效性.A comprehensive analysis of the characteristics of the network lateral and vertical grey incidence analyses is given,and then a new pruning algorithm based on grey incidence analysis for high-order neural network is presented. In this algorithm,grey incidence analysis is used to compare the correlation degree of each output sequence of the network units. The pruned connections and the incorporated connections are determined by the network vertical and lateral grey incidence analyses,respectively. The topological structure of the network is adjusted dynamically. After training,the pruned network has the optimum topological structure and obtains good generalization. The simulation results show the rationality and effectiveness of the proposed approach.

关 键 词：灰色关联分析 高阶神经网络 灰色关联度 剪枝算法 最小二乘法 

分 类 号：TP183[自动化与计算机技术—控制理论与控制工程]