摘要
大多统计模型的输出与输入都是高度非线性和线性相叠加的关系,为了更好地实现数据驱动的研究,本文提出了一种隐含层组合型的ELM(Extreme Learning Machine with Hybrid Hidden Layer,HHL-ELM)神经网络。该HHL-ELM神经网络在传统的ELM网络的隐含层中增加一个特殊的节点,该特殊节点的激活函数与隐含层其他节点激活函数不同,从而形成了一种隐含层组合的网络结构,试图增强ELM网络模型的输出。同时,本文利用UCI标准数据集中的Housing数据集进行了测试,并通过工业应用实例进行了验证。最后进行了模型对比,结果表明HHL-ELM网络在处理复杂数据时具有精度高的特点,为神经网络发展及其应用提供了新思路。
In most statistical models, the output and input have the relationship of highly nonlinear and linear superposition. In order to achieve better data-driven research, in this paper Extreme Learning Machine with Hybrid Hidden Layer （HHL-ELM） is proposed. Compared with the traditional ELM, some special neurons are added in the hidden layer of HHL-ELM. The activation functions in the special neurons are different from those in the other hidden neurons, which form a kind of network structure with hybrid hidden layer that can enhance the ability of the output of ELM. Meanwhile, the Housing data set from UCI standard data set is selected to exam the HHL-ELM model. Meantime, the HHL-ELM model is applied to model chemical processes. Compared with the ELM, the results show that HHL-ELM has a higher accuracy when dealing with complex data sets, which provide a new idea for the development of neural network.
出处
《计算机与应用化学》
CAS
CSCD
北大核心
2013年第12期1393-1396,共4页
Computers and Applied Chemistry
基金
国家自然科学基金项目(61074153)
关键词
极限学习机
建模
复杂数据
extreme learning machine, modeling, complex datasets
