摘要
针对深度信念网(Deep belief network,DBN)学习连续数据时预测精度较差问题,提出一种双隐层连续型深度信念网.该网络首先对输入数据进行无监督训练,利用连续型传递函数实现数据特征提取,设计基于对比分歧算法的权值训练方法,并通过误差反传对隐层权值进行局部寻优,给出稳定性分析,保证训练输出结果稳定在规定区域.利用Lorenz混沌序列、CATS序列和大气CO_2预测实验对该网络进行测试,结果表明,连续型深度信念网具有结构精简、收敛速度快、预测精度高等优点.
A continuous deep belief network(c DBN) with two hidden layers is proposed to solve the problem of low accuracy of traditional DBN in modeling continuous data. The whole process is to train the input data in an unsupervised way using continuous version of transfer function, to design the contrastive divergence in hidden-layer training process,and then to fine-tune the net by back propagation. Besides, hyper-parameters are analyzed according to stability analysis,as is given in the paper, to make sure the network finds the optimal. Experiments on Lorenz, CATS benchmark simulation and CO2 forecasting show a simplified structure, fast convergence speed and accuracy of this c DBN.
出处
《自动化学报》
EI
CSCD
北大核心
2015年第12期2138-2146,共9页
Acta Automatica Sinica
基金
国家自然科学基金(61203099
61225016
61533002)
北京市科技计划课题(Z141100001414005
Z141101004414058)
高等学校博士学科点专项科研基金资助课题(20131103110016)
北京市科技新星计划(Z131104000413007)
北京市教育委员会科研计划项目(KZ201410005002
km201410005001)资助~~
关键词
深度学习
神经网络
结构设计
稳定分析
时序预测
Deep learning
neural networks
structural design
stability
time series forecasting