基于并行BP神经网络的路由查找算法_胥小波.pdf

第 33 卷第 2 期 通 信 学 报 Vol.33 No. 2 2012 年 2 月 Journal on Communications February 2012 基于并行BP 神经网络的路由查找算法 1,2 1,2 1,2 1,2 1,2 胥小波 ，郑康锋 ，李丹 ，杨义先 ，钮心忻 （1. 北京邮电大学 信息安全中心，北京 100876 ；2.北京邮电大学 灾备技术国家工程实验室，北京 100876) 摘 要：结合 Bloom-filter 算法和并行反向传播神经网络，提出了一种新的基于并行神经网络的路由查找算法 (BFBP) 。该算法满足路由查找的需求，只需学习路由条目的网络ID，且易于扩展到 IPv6 地址查询。研究结果表 明，相比于己有的神经网络路由查找方法，该算法需要学习的条目数平均减少了520 倍，提高了学习效率，为神 经网络应用于路由查找创造了有利条件。 关键词：神经网络；路由查找；反向传播；Bloom-filter 算法 中图分类号：TP393.03 文献标识码：A 文章编号：1000-436X(2012)02-0061-08 Routing lookup algorithm based on parallel BP neural network XU Xiao-bo1,2, ZHENG Kang-feng1,2, LI Dan1,2, YANG Yi-xian1,2, NIU Xin-xin1,2 (1. Information Security Center, Beijing University of Posts and Telecommunications, Beijing 100876, China; 2.National Engineering Laboratory for Disaster Backup and Recovery, Beijing University of Posts and Telecommunications, Beijing 100876, China) Abstract: A new routing lookup algorithm based on Bloom-filter algorithm and parallel back-propagation neural net- works (BFBP) was proposed. The algorithm could meet the challenges of routing lookup and just had to learn the net- work ID moreover, it was equally attractive for IPv6. The results show that compared to other routing lookup methods based on neural network, BFBP algorithm reduces the average number of items which neural network has to learn by 520 times, improves learning efficiency of neural networks, and creates favorable conditions for the application of neural network in the area of routing lookup.