基于CRBM算法的时间序列预测模型研究

doi:10.3785/j.issn.1008-9497.2016.04.011

摘要
图/表
参考文献
相关文章 (0)

全文: PDF (4945 KB) HTML( )
输出: BibTeX | EndNote (RIS)

摘要针对受限玻尔兹曼机（restricted Boltzmann machines，RBM）算法对时序数据预测存在抽取抽象特征向量能力较差和梯度下降能力有限的问题，基于CRBM（conditional restricted Boltzmann machines）算法以及信念网络（deep belief network，DBN）模型，构建了一种非线性的CRBM-DBN深度学习模型，并采用高斯分布处理输入特征值和对比散度抽样，用于预测时序数据.实验以浙江省近岸海域赤潮时序数据作为输入特征值，讨论该模型的深度及参数选取，并与经典的深度学习模型RBM、DAE和浅层学习中的BP神经网络进行对比，实验验证CRBM对于赤潮时序数据的预测拟合度要明显优于其他3种模型，该模型可有效用于赤潮类时序数据的趋势性预测.

	服务

	把本文推荐给朋友
	加入我的书架
	加入引用管理器
	E-mail Alert
	RSS
	作者相关文章
	周晓莉
	张丰
	杜震洪
	曹敏杰
	刘仁义

关键词 ：受限玻尔兹曼机, CRBM深度建模, 深度信念网络模型, 高斯分布

Abstract：Restricted Boltzmann machines (RBM) algorithm has a poor performance in extracting feature vector and gradient descent when it is used to predict time-series data. To solve the above problems, a non-linear deep learning model was constructed based on conditional restricted Boltzmann machines (CRBM) combining with deep belief network (DBN). The model processed the input feature vectors with Gaussian distribution and samples with classical contrastive divergence to predict continuous time-series data. Our experiment adopted the time-series data of red tide in Zhejiang costal, and discussed the selection of network depth and training parameters in the model, then compared the deep learning model to classical RBM, DAE deep learning network and BP neural network shadow learning. The results showed that the prediction fitting of CRBM was superior to the other three models. This model can effectively predict the time-series of red tide.

Key words： restricted Boltzmann machines deep architecture of CRBM deep belief network model Gaussian distribution

收稿日期: 2015-08-06

ZTFLH:

P208

基金资助:国家自然科学基金资助项目（41471313，41101356，41101371，41171321）；国家科技基础性工作专项（2012FY112300）；海洋公益性行业科研专项经费资助（2015418003，201305012）.

通讯作者: 杜震洪,ORCID:http://orcid.org/0000-0001-9449-0415,E-mail:duzhenhong@zju.edu.cn. E-mail: duzhenhong@zju.edu.cn

作者简介: 周晓莉(1990-),ORCID:http://orcid.org/0000-0002-9097-5682,女,硕士研究生,主要从事海洋GIS及深度学习研究.

引用本文:

周晓莉, 张丰, 杜震洪, 曹敏杰, 刘仁义. 基于CRBM算法的时间序列预测模型研究[J]. 浙江大学学报（理学版）, 2016, 43(4): 442-451.
ZHOU Xiaoli, ZHANG Feng, DU Zhenhong, CAO Minjie, LIU Renyi. A study on time series prediction model based on CRBM algorithm. Journal of ZheJIang University(Science Edition), 2016, 43(4): 442-451.

链接本文:

http://www.zjujournals.com/sci/CN/10.3785/j.issn.1008-9497.2016.04.011 或 http://www.zjujournals.com/sci/CN/Y2016/V43/I4/442

[1] HINTON G E, SALAKHUTDINOV R R. Reducing the dimensionality of data with neural networks[J]. Science,2006,313(5786):504-507.
[2] KUREMOTO T, KIMURA S, KOBAYASHI K, et al. Time series forecasting using a deep belief network with restricted Boltzmann machines[J]. Neurocomputing,2014,137:47-56.
[3] SALAKHUTDINOV R, MNIH A, HINTON G. Restricted Boltzmann machines for collaborative filtering[C]// Proceedings of the 24th International Conference on Machine Learning.New York:ACM,2007:791-798.
[4] MARKOFF J. Scientists see promise in deep-learning programs[N]. New York Times,2012-11-27.
[5] LAROCHELLE H, BENGIO Y. Classification using discriminative restricted Boltzmann machines[C]// Proceedings of the 25th International Conference on Machine Learning.New York:ACM,2008:536-543.
[6] MNIH V, LAROCHELLE H, HINTON G E. Conditional restricted Boltzmann machines for structured output prediction[C]// 28th Conference on Uncertainty in Artificial Intelligence,2012:514-522.
[7] LÄNGKVIST M, KARLSSON L, LOUTFI A. A review of unsupervised feature learning and deep learning for time-series modeling[J]. Pattern Recognition Letters,2014,42(1):11-24.
[8] 陈翠平.基于深度信念网络的文本分类算法[J].计算机系统应用,2015,24(2):121-126. CHEN Cuiping. Text categorization based on deep belief network[J]. Computer Systems and Applications,2015,24(2):121-126.
[9] 刘建伟,刘媛,罗雄麟.玻尔兹曼机研究进展[J].计算机研究与发展,2014,51(1):1-16. LIU Jianwei, LIU Yuan, LUO Xionglin. Research and development on Boltzmann machine[J]. Journal of Computer Research and Development,2014,51(1):1-16.
[10] HINTON G E. A practical guide to training restricted Boltzmann machines[C]//Neural Networks: Tricks of the Trade. Heidelberg:Springer,2012:599-619.
[11] SPILIOPOULOU A. Investigation of Deep CRBM Networks in modeling Sequential Data[D]. Edinburgh:University of Edinburgh,2008.
[12] MONTUFAR G, AY N, ZAHEDI K. Geometry expressive power of conditional restricted Boltzmann machines[R]. New Mexico:Santa Fe Institute,2014.
[13] BENGIO Y. Learning deep architectures for AI[J]. Foundations and Trends in Machine Learning,2009,2(1):1-127.
[14] CHEN S, AMMAR H B, TUYLS K, et al. Conditional restricted Boltzmann machines for negotiations in highly competitive and complex domains[C]// Proceedings of the Twenty-Third International Joint Conference on Artificial Intelligence.Palo Alto:AAAI Press,2013:69-75.
[15] TAYLOR G W, HINTON G E. Factored conditional restricted Boltzmann machines for modeling motion style[C]// Proceedings of the 26th Annual International Conference on Machine Learning.New York:ACM,2009:1025-1032.
[16] SALAKHUTDINOV R, LAROCHELLE H. Efficient learning of deep Boltzmann machines[J/OL]. Journal of Machine Learning Research,2010,9(8):693-700.
[17] HINTON G E. To recognize shapes, first learn to generate images[J]. Progress in Brain Research,2007,165(6):535-547.
[18] SCHÖLKOPF B, PLATT J, HOFMANN T. Modeling human motion using binary latent variables[C]//Advances in Neural Information Processing Systems 19: Proceeding of the 2006 Conference. Cambridge: MIT Press,2007:1345-1352.
[19] 单外平.基于深度信念网络的变速器故障分类识别研究[D].广州:华南理工大学,2015. SHAN Waiping. Classification and Recognition of Transmission Fault Based on Deep Belief Network[D]. Guangzhou: South China University of Technology,2015.