双并行结构优化的极限学习机及其在回归数据集中的应用  

An Enhanced Extreme Learning Machine with a Double Parallel Structure(Dp-Elm) and Its Application to Regression Datasets

作　　者：韩静文[1] 崔玉龙[1] 辛博[2] 张雅静[1] 

机构地区：[1]北京化工大学信息科学与技术学院,北京100029 [2]北京国电智深控制技术有限公司,北京102200

出　　处：《内蒙古大学学报（自然科学版）》2017年第2期181-188,共8页Journal of Inner Mongolia University：Natural Science Edition

摘　　要：极限学习机(ELM)发展自单隐含层前馈神经网络算法,其理论简单,运行快速,应用非常广泛.为了提高ELM的泛化性能,提出了一种带有双并行结构的优化ELM算法(DPELM).在DP-ELM中,建立输入层和输出层间特殊的连接,使得DP-ELM的输出节点不仅可以接收隐含层节点的信息,也可直接接收输入节点的自信息.利用3组回归数据集验证算法性能,实验结果证明,与ELM相比,DP-ELM可以达到更好的回归精度以及更稳定的泛化能力.Extreme learning machine（ELM）, a learning algorithm for single-hidden layer feed- forward neural nctworks（SLFNNs）, has been successfully applied in many fields and is simple in theory and fast in implementation. In order to enhance the generalization performance of ELM, an enhanced ELM is proposed with a double parallel structure（DP-ELM）. In DP-ELM,a special channel is established to directly connect the input layer and the output layer. Then the output nodes of DP-ELM can receive the information from the hidden layer nodes and also receive the self-information directly from the input nodes. To test the validity of DP-ELM,it is applied to 3 regression applications. Experimental results show that, compared with ELM,DP-ELM can achieve much better regression precision and have better stable ability in the generalization phase.

关 键 词：极限学习机 单隐含层前馈神经网络 神经网络 回归 

分 类 号：TP181[自动化与计算机技术—控制理论与控制工程]