一种基于蚁群算法与粗糙集的混合BP神经网络  

A Mixed BP Neural Network Based on Ant Colony Algorithm and Rough Set

作　　者：洪月华[1] 

机构地区：[1]广西经济管理干部学院计算机系,广西南宁530007

出　　处：《微电子学与计算机》2014年第4期156-159,共4页Microelectronics & Computer

基　　金：广西自然科学基金青年项目(2012GXNSFBA053178);广西混杂计算与集成电路设计分析重点实验室开放基金课题(2012HCIC05)

摘　　要：为了解决BP神经网络对高维冗余样本分类时收敛速度慢、易陷入局部极小值问题,提出基于蚁群算法与粗糙集的混合BP神经网络分类模型.该混合BP神经网络用粗糙集对样本进行约简和降维,输入层神经元个数得到减少,降低了训练神经网络的计算复杂度,用蚁群算法解决了选取神经网络权值和阈值的随机性,避免了因其而导致的易陷入局部极小值的不足.对UCI数据库中数据集的测试结果说明,提出的混合BP神经网络对高维冗余复杂样本进行分类是可行的,性能远远比传统BP神经网络和蚁群神经网络优越.In order to solve the problem that when BP neural network classify high dimensional redundant sample the convergence speed is slow and is easy to fall in local minimum problem, a mixed BP neural network model based on ant colony algorithm and rough set is proposed. In the mixed BP neural network, rough set is used to reductive and reduce the dimension of sample, so the input layer neuron number is reduced, which reduce the computational complexity of the training of the neural network, and ant colony algorithm has solved the random of selection neural network weights and threshold, which has avoided the problem of being easy to fall in local minimum problem. The results gotten by testing data sets in the UCI database, the mixed BP neural network is feasible to classify high dimensional redundant sample, and the performance is more better than the traditional BP neural network and ant colony neural network.

关 键 词：蚁群算法ACA 粗糙集 BP神经网络 

分 类 号：TP183[自动化与计算机技术—控制理论与控制工程]