基于禁忌搜索的前向神经网络在函数逼近中的应用  

Forward Neural Network Based on Tabu Search and Its Application in Function Approximation

作　　者：贺一[1] 

机构地区：[1]西南师范大学计算机与信息科学学院

出　　处：《西南师范大学学报（自然科学版）》2004年第3期361-365,共5页Journal of Southwest China Normal University(Natural Science Edition)

基　　金：教育部科技重点项目(2000114;104262);重庆市科委基金项目(2003-7881);重庆师范大学校级课题.

摘　　要：为改善前向神经网络的性能,将禁忌搜索作为前向神经网络的训练算法,采用了一种集中性与多样性的自适应搜索策略,以提高禁忌搜索的有效性.并以正弦函数和sinc函数的逼近为例,验证了算法的可行性和有效性.To improve the performance of FNN, this paper attempted to employ a new global optimizing algorithm——Tabu Search (TS) in FNN for its training, and a novel adaptive search strategy of intensification and diversification was proposed to improve efficiency of TS. Taking function approximation as samples, many simulating experiments were carried out. The result shows: TS is much better than BP algorithm in function approximation, especially in the functions that their nonlinear degrees are very higher.

关 键 词：禁忌搜索 前向神经网络 函数逼近 反向传播算法 

分 类 号：TP181[自动化与计算机技术—控制理论与控制工程]