多子群粒子群集成神经网络  

Artificial neural network ensemble based on multi-sub-swarm particle swarm optimization

作　　者：潘希姣[1] 

机构地区：[1]安徽交通职业技术学院汽车与机械工程系,合肥230051

出　　处：《安徽建筑工业学院学报（自然科学版）》2007年第2期38-40,共3页Journal of Anhui Institute of Architecture(Natural Science)

摘　　要：为了使参加神经网络集成的个体差异度较大,从而提高网络集成的泛化能力,本文提出一种新的基于多子群粒子群算法的神经网络集成方法。每个子群通过补充差异度独立训练出一批神经网络,从每个子群中选择一个最优个体参加网络集成,实验使用了UCI标准数据集。实验证明,该算法的识别能力要好于Boosting、Bagging等传统方法。To improve the generalization capability of neural network ensemble and enlarge the diversity of every individual artificial neural network （ANN）, a novel ANN ensembles based on multi-sub-swarm particle swarm optimization were proposed. Each sub-swarm trains different ANN through complementary diversity. The best individual was chosen from each sub-swarm to take part in the ensemble. The experiments used UCI benchmark datasets, the experimental results show that the recognition rate of this algorithm is better than the Boosting and Bagging algorithm.

关 键 词：神经网络集成 粒子群优化算法 

分 类 号：TP183[自动化与计算机技术—控制理论与控制工程]